Technology as a Competitive Advantage: How AI Changes Work and the Skills We Need

Cover image: AI-generated conceptual image of lifelong learning with fictional people.

AI can draft a text, organize information or suggest a solution in seconds. That moves a task forward. Whether it produces good work depends on what happens next: can someone judge whether the result is correct, addresses the problem and can responsibly be used in that particular setting?

This question connects my work in technology, education and business practice. The occasion for this article is my contribution, “Technology as a Competitive Factor,” at NextWork Bavaria in Munich on October 29, 2026. Its program brings together technology, organization and people. For me, their interaction is where the important work begins.

I find little value in reducing the future of work to a bet on how many people will eventually still be needed. I want to understand what people will do, how they will work and how we can best prepare them. AI and automation can contribute a great deal. We need to design their use with the same care that we give to educating the people who will work with them.

My argument is that technology becomes a competitive advantage when people can use it well. That requires professional competence, a willingness to learn and organizations that develop both.

As of October 29, 2026. The article distinguishes research findings, demographic scenarios and my own conclusions where each is discussed.

How AI Changes Tasks and Occupations

A job contains many different activities. A professional searches for information, talks to other people, investigates a problem, documents the work and makes decisions. Technology can support or automate these activities to different degrees. Claims about replacing an occupation therefore deserve a closer look at the work involved.

The term AI also covers different tools. A language model can help with writing. A system trained on measurement data can flag anomalies. A robot can perform physical tasks within an appropriately designed working environment. These applications place very different demands on data, integration, safety and skills.

Technical feasibility alone tells us little about whether a solution will operate reliably and economically in everyday use. It must fit existing processes. Someone has to supply suitable data, establish access permissions, manage exceptions and determine whether the whole process has improved.

The ILO’s methodological brief on workers’ exposure to AI explains why measures of technical exposure do not, on their own, predict actual job losses or productivity gains. They initially describe which tasks could be affected.

This suggests a useful sequence for practice. Understand the task and its constraints, decide what technology should contribute, then assess which activities disappear, which emerge and which skills they require. I also discuss this attention to real processes in my article on robotics and automation at EDAY in Vienna.

What We Know About Productivity and Employment

There are good reasons for optimism. The history of automation shows how technology can complement human work and generate new demand. David Autor examines these mechanisms in his essay on the history and future of workplace automation. Technology also changes the activities employers need and what they pay for them. A positive employment balance is not automatic.

Negative effects have been documented too. Acemoglu and Restrepo’s study of industrial robots in US labor markets finds employment and wage losses in more exposed local markets. That finding concerns a particular technology and research setting. It is no more a universal rule than the claim that every new technology necessarily creates more jobs than it displaces.

Productivity gains can also be measured. In “Generative AI at Work,” a study of 5,172 customer support workers, AI assistance increased issues resolved per hour by an average of 15 percent. Less experienced or lower-skilled workers benefited particularly strongly. This is a finding from that application, rather than a general performance multiplier for every occupation.

My practical conclusion is to look for improvements where AI assistance helps a specific task, then examine the outcome across the entire process. A faster draft can be useful. If checking it takes longer than the work saved, or errors become expensive later, the calculation changes.

Productivity therefore means more to me than speed. Quality, reliability, resource use and the ability to handle exceptions matter too. Counting generated documents or interactions with a tool tells us little about whether a business is doing better work.

Why Demographic Change Adds Urgency

For Germany, Destatis projects 3.2 million fewer people aged 20–66 under high net immigration, or up to 4.9 million fewer under low net immigration, by the mid-2030s, relative to its baseline.

These population scenarios do not estimate unfilled jobs. Participation, migration, skills, working hours and demand influence the actual labor market.

The international picture also varies. The ILO’s Employment and Social Trends 2026 report distinguishes aging economies from countries with rapidly expanding labor supply that need to create enough good jobs. This does not imply a uniform global labor shortage beginning in a particular year.

For a business in Germany, the question remains pressing: how can it maintain its capabilities as experienced people leave and new recruits are difficult to find? Technology can help make better use of scarce working time and improve access to knowledge. People also need a realistic opportunity to develop the required skills. Recruitment difficulties and problems entering the labor market can coexist.

An Example: Troubleshooting With AI Assistance

Consider a hypothetical example in a technical workplace. A machine develops a fault. An AI assistant could search approved manuals, summarize earlier records and suggest possible causes. This might make the preparatory work easier.

A qualified professional still has to assess whether a proposed explanation fits the actual situation. Are the readings plausible? Does the instruction apply to the correct machine configuration? Which safety requirements apply? What testing is permitted, and when should someone else become involved?

Step	Possible AI contribution	Human responsibility
Define the problem	Search documentation and organize relevant information	Describe observations, operating conditions and constraints accurately
Investigate causes	Suggest hypotheses and relevant references	Check measurements and reject unsuitable suggestions
Choose an approach	Make relevant instructions accessible	Assess safety, responsibilities and permitted actions
Document the outcome	Structure a draft report	Record actual actions and findings in a traceable way

This example does not describe a client project or claim a measured saving. It illustrates the questions a company should answer before turning a technical possibility into a new workflow.

The training implications are particularly interesting. An early-career employee could first propose an explanation independently, then examine the AI suggestion and discuss the decision with an experienced colleague. That would combine tool use with the development of professional judgment. Learning becomes visible when the person can explain why a suggestion is appropriate or needs to be rejected.

Having an answer available does not replace that ability. A well-written fault report alone does not make someone an experienced maintenance professional. A tool can support a learning pathway. We have to design the pathway ourselves.

The Skills That Matter in an AI-Enabled Workplace

I would organize the discussion about future skills around six connected areas. This is a practical framework, rather than a permanent ranking. The OECD synthesis “Skills in the AI age” emphasizes strong foundations alongside digital and complementary skills. The depth of technical knowledge required depends on the work.

Domain knowledge and foundations. Understand relationships and recognize significant errors.
Problem and process understanding. Describe goals, constraints and interfaces.
AI and data literacy. Use suitable tools and account for their limits.
Judgment and responsibility. Evaluate outputs and justify decisions.
Collaboration and communication. Share knowledge and connect different perspectives.
Learning and independence. Explore new approaches and develop the ability to act competently.

Domain Knowledge Provides the Foundation

Asking a useful question often requires understanding the subject already. Assessing a technical proposal requires professional standards. Using a statistical claim responsibly requires an understanding of scale and uncertainty. I would therefore strengthen educational foundations and make their connection to practical application more explicit.

In the troubleshooting example, this means understanding how the system works and which observation makes a particular cause plausible. That knowledge enables a check that goes beyond the impression that the text sounds convincing.

Understand Problems and Processes

Technology is easier to use well when the purpose is clear. What should improve? Where does the task start and end? Which handovers to other people or systems are involved?

These questions connect technical and organizational skills. A locally faster solution can create extra work at the next interface. People should be able to recognize such relationships and explain their requirements clearly.

Learn to Use AI Appropriately

Not everyone needs to develop a model or write code. In many roles, the initial priorities are choosing a suitable tool, providing relevant context and evaluating the result. This includes recognizing which information may be used and when a different approach would be more appropriate.

Prompting belongs within this practical competence. Designing a good workflow also involves data, interfaces, quality and responsibility. Specialist technical roles need the corresponding depth of expertise.

Practice Judgment

A valuable learning task might involve spotting a plausible error. Why is the proposed calculation unsuitable? Which assumption is missing? What source needs checking?

I would deliberately include such exercises in education and professional development. What matters is the reasoned decision. A checkmark beneath an automatically generated answer is meaningful only if the person reviewing it can understand and challenge it.

Make Collaboration Understandable

New tools often affect several people’s work. Business functions, technology teams, management and people involved in the process need to communicate. That calls for understandable explanations and a willingness to take questions from another perspective seriously.

My proposal for an understandable one-page explanation of publicly funded technology in the New York and UN article follows the same conviction: people need to understand an issue if they are to help shape decisions about it. Inside a company, that principle can be applied to the goals, limits and responsibilities of an AI deployment.

Enable Learning and Preserve Independence

For me, a willingness to learn means trying new technologies, asking questions and developing one’s way of working. It also includes giving a reasoned explanation for stopping an experiment or declining to use a tool for a particular task.

This attitude is not determined by someone’s year of birth. My article “AI After 60” also addresses curiosity, critical evaluation and the opportunity to form an informed opinion. Professional development should offer the same access. People should also retain enough understanding of their work to respond sensibly when an output is unusable or a system fails.

Who Will Develop Tomorrow’s Experts?

In conversations with people in the market, I have noticed a more sober view of the idea that AI could simply make junior employees unnecessary. That personal impression is not a labor-market statistic. My case for strong early-career training does not depend on claiming that hiring has undergone a general reversal.

The current Stanford analysis “Canaries in the Coal Mine?” continues to show less favorable employment trends for 22–25-year-olds in AI-exposed US occupations than for less-exposed peers, primarily through reduced hiring. The authors describe the patterns as descriptive; they do not establish a definitive causal AI effect or describe all junior positions in Germany.

Organizations that need experienced professionals in a few years must create opportunities to learn today. As entry-level tasks change, the development of experience has to change with them. Automating tasks and subsequently lamenting a lack of experience is an inadequate approach.

I would assess competence-based education by what a person can do professionally and explain. This might be a justified diagnosis, a traceable analysis or a responsibly completed practical task. Such abilities require knowledge, practice and feedback. A course certificate or an impressive output alone does not tell us enough.

AI can be used within that learning: to offer alternative explanations, compare ways of solving a problem or provide exercises. I would use it at suitable points throughout a program. Learners should also find errors, encounter limits and explain their own decisions.

Mentors have a demanding role. They need to distinguish understanding from the ability to reproduce a plausible output. Follow-up questions, changed conditions and tasks that reveal the reasoning process can help. A work sample can certainly involve AI. The person should then be able to defend the decision and respond to new information.

Learning also continues beyond a first qualification. An experienced employee brings valuable knowledge and may need support with a new tool. A new recruit may be familiar with the interface while still developing professional depth. Both need appropriate opportunities to learn. The label “digital” tells us little about a complete competence profile.

The Responsibilities of Employers and Leaders

An employer can ask people to be willing to learn. Without time and appropriate conditions, that request may achieve little. I therefore consider competence development part of the investment in technology.

When introducing a system, an organization should establish who will perform which tasks and how those people will be prepared. People who do the work every day can offer important insights into exceptions, unnecessary handovers and practical obstacles. Their knowledge belongs in the design from an early stage.

The economics deserve a complete view too. Integration, operation, checking, error handling and training all belong alongside the tool itself. That applies to an AI assistant and to an automation system involving physical hardware. The purchase alone tells us little about the eventual benefit.

For a limited pilot, I would agree on a few understandable measures of success. Is the task performed reliably? Does quality improve? What effort does the whole process require? Can the people involved handle exceptions? How do they experience their work?

The use of any time saved also needs discussion. A company could provide more output, improve quality, relieve pressure on employees or develop new activities. The appropriate combination deserves a deliberate decision. Otherwise, a technical improvement could become another source of intensified work.

Responsible design also requires clear responsibilities and a way to report problems. My thinking on Robotic & AI Governance starts here: a system’s capabilities, the limits of its use and human responsibility need to fit together.

What Human in the Loop Should Mean in Practice

People remain important in many processes for setting goals, making decisions, operating systems and taking responsibility. Merely having a person present does not make a process well controlled. I would check four conditions: does the person understand the task, have the relevant information, have enough time, and have both the authority and ability to intervene?

If a condition is missing, oversight can become a formality. Someone expected simply to pass an output along without the expertise to assess it has not been given a meaningful control role.

This does not mean that every automated step needs manual approval. Human involvement should fit the task and its consequences. Nor can we calculate total job creation merely from the existence of maintenance and oversight work.

My optimism concerns what well-qualified people can achieve with suitable tools. AI can offer substantial assistance to experts. How particular tasks and positions change nevertheless remains a specific economic and organizational question. A blanket employment guarantee would be poor advice.

Where Organizations and Employees Can Start

A useful starting point is a manageable assignment. I would select an existing process and work through these steps together:

Understand the task. Describe the problem, the desired outcome and current difficulties.
Define technology’s contribution. Identify suitable assistance, limits and necessary human decisions.
Develop competence. Plan responsibilities, learning time, practice and professional feedback.
Run a limited trial. Examine quality, total effort and working conditions against agreed criteria.
Evaluate and decide. Improve the process, expand it or explain why the trial should end.

Employees can apply the same questions to their own work. Which recurring activity takes time? How do they recognize a good result? What knowledge do they need to check it? A small trial agreed with the organization may teach more than a collection of general predictions about the future.

For people entering a profession, my advice is to combine familiarity with tools and depth of subject knowledge. Experiment, verify the result, explain your approach and seek feedback. These activities make the development of professional competence visible.

The Future of Work Requires Education and Deliberate Design

I expect people to continue playing a central role in work as new technologies emerge. How well they can fulfill that role depends on our decisions. The name of the next technology changes little about the responsibility to prepare people to use it well.

That is why I want lifelong learning to mean something concrete: time to practice, access to appropriate tools, strong foundations and feedback from people who understand the subject. For me, competence-based education translates these conditions into professional action that can be demonstrated and explained.

Employers, educational institutions and employees each have a part to play. I want us to use the technical possibilities while preserving the ability to understand, decide and take responsibility. That is where I see technology becoming a competitive advantage.

Frequently Asked Questions About AI and the Future of Work

What does the future of work mean in the context of AI?

Here, the future of work refers to changes in tasks, collaboration, skills and working conditions brought about by new technologies. For organizations, it involves designing specific processes. For employees, it concerns the activities they will perform and the abilities they need to develop. The term does not describe a single predetermined future.

Will AI replace people and their jobs?

AI can take over activities and consequently change or displace positions. It can also support human work and enable new tasks. Outcomes depend on the application, economic demand and organizational decisions. A prediction of technical capability is therefore not a measured number of future job losses. The research sources in the article explain this distinction.

Which skills matter when working with AI?

I would develop domain knowledge, problem understanding, AI and data literacy, judgment, collaboration and learning ability together. The appropriate combination depends on the role. A useful learning task combines tool use with professional evaluation: the person should be able to explain what they accepted, changed or rejected, and why.

Will everyone need to code or become an AI expert?

Many roles require practical competence suited to the task. People should be able to use appropriate tools, evaluate results and recognize limits. Those developing, integrating or safely operating systems need greater technical depth. AI’s broad relevance does not imply the same educational objective for every occupation.

Why should companies continue to train junior employees and apprentices?

Experience develops through learning and supervised practice. Organizations that will need experienced professionals should make that development possible. Changing entry-level tasks require adapted learning pathways, including independent attempts, AI-assisted comparisons and professional feedback. This is a strategic argument for education and staff development, rather than a guarantee about hiring in every occupation.

What does competence-based education mean in practice?

It focuses on whether a person can perform a professional task competently and explain the decisions involved. Knowledge, practice and feedback are essential. One possible assessment is a work sample followed by questions and changed conditions. AI can be part of the exercise; assessment should also examine personal understanding and the ability to act.

How can employers support lifelong learning?

I recommend dependable learning time, suitable tools, realistic tasks and accessible professional guidance. Development should connect to actual work and accommodate different starting points. A culture that welcomes questions and reasoned objections matters too. These arrangements make shared responsibility for learning practical for both employers and employees.

What does human in the loop mean for AI at work?

The term describes human participation in a partly automated process. It is meaningful when someone understands the task, has relevant information and sufficient time, and can actually intervene. The necessary involvement depends on the activity. A formal confirmation without an independent assessment does not establish reliable control.

Will AI solve Germany’s shortage of skilled workers?

AI can help make more effective use of available working time. It does not automatically solve problems involving skills, access or staffing. A shrinking working-age population increases pressure to act, but does not directly predict unfilled positions. Training, good working conditions and appropriate process design remain responsibilities in their own right.

How can I start using AI sensibly in my area of work?

Choose a clearly bounded task and agree on the trial with the people responsible. Define a good result, suitable data and necessary checks. Compare quality and total effort with the existing approach. Record what you learned and which abilities you still need for reliable use.

Sources and Further Reading

One Page That Gives People a Say: My Proposal at the UN

Cover image: AI-generated illustration of dialogue about technology with fictional people; not a photograph of the event.

Anyone receiving public funding to research or develop new technology should explain to the public, in understandable language, what they are working on. For me, one page would be the minimum. What problem is the technology meant to solve? What can it demonstrably do, where are its limits, and how can the people affected have a say?

I have been making this proposal for years. At the UN Digital Cooperation Day in New York, I raised it again. For me, it captures a practical task of Robotic & AI Governance: people need to be able to assess technology and express what they want while important decisions are still open.

An understandable text is the starting point. Participation begins when questions and objections receive an accountable response: who made the decision, and what changed as a result of the feedback?

In this article: first, the documented outcomes from New York and the original recordings; then my personal commentary and a practical proposal for publicly funded technology.

Go to the outcomes · Go to my one-page proposal · Go to the questions and answers

Event: 21 September 2026 · Research and source review: 18 October 2026

Part 1 · Report: What the UN Digital Cooperation Day Produced

On 21 September 2026, the UN Office for Digital and Emerging Technologies (ODET) held the second UN Digital Cooperation Day in New York. During the high-level week of the UN General Assembly, the event brought together three tasks: assessing scientific evidence, informing policy decisions, and building the capacity for wider participation in AI. The official programme documents the sessions and recordings.

The outcomes at a glance

Topic	What is available	Status and significance
National AI strategies	Introduction of the UN Blueprint for National AI and Data Strategies, drawing on experience from Lesotho and Guinea	A practical guide to implementing national strategies. Lesotho had already piloted it in August.
Scientific risk assessment	The first thematic brief from the Independent International Scientific Panel on AI	Released on 21 September as an advance unedited version; its focus is AI agents, misalignment and risks of losing human control.
AI and disaster risk reduction	A new partnership between ODET and the UN Office for Disaster Risk Reduction (UNDRR)	An announced collaboration on using AI to reduce disaster risk.
Work, open AI and cross-border testing	Discussions of skills, open-source AI and shared sandboxes for AI governance	Topics for exchange; a session alone does not amount to an adopted international standard.

Sources for the overview: UN event report, ODET report on Lesotho's pilot, and the AI Panel's first thematic brief.

National AI strategies: from policy ambition to practical planning

The Blueprint starts with a practical question: how can a national AI strategy become something a country can actually implement? In Lesotho, the work covered computing capacity, energy, connectivity, data, skills and financing. The August workshop produced a draft national roadmap for computing capacity. The New York launch brought together Lesotho's experience and Guinea as the next country to apply the approach. The ODET report describes that process.

AI agents: a scientific brief on risks and questions of control

The new thematic brief examines how the behaviour of capable AI agents can diverge from human intentions. It takes a documented incident during the safety evaluation of such systems as its starting point. The Panel places the incident in its research context and considers safeguards used in other fields with demanding safety requirements.

The limits of the assessment matter: the brief estimates neither the probability nor the timing of a severe future loss of control. It does not prescribe binding policy measures. The original publication is labelled an “advance unedited version”.

The preliminary overall report, released on 1 July 2026, and the new thematic brief serve different purposes. The overall report assesses the field more broadly; the September brief explores a particular risk question. The scientific Panel's FAQ explain this sequence of publications. My earlier assessment of the preliminary UN AI report remains available as a contribution written at that time.

Work, education and a shared evidence base

The session on the future of work focused on education, skills and the readiness of people and institutions. This included the question of who can participate in an AI-enabled economy. Recording: work and education.

The “Panel of Panels” compared international scientific assessments: where do they agree, where are the gaps in knowledge, and which perspectives are missing? Experiences from the Global South formed part of that discussion. The focus was on improving the evidence available for governance. Recording: Panel of Panels.

What can be recorded as an outcome

The day brought together a practical guide, a new scientific publication and an announced partnership. Other sessions provided opportunities to exchange experience on implementation and cooperation. The official event report places these steps in the context of implementing the Global Digital Compact. It does not document a decision adopting my one-page proposal.

Watch the official recordings

The UN programme links to three parallel recordings. The timestamps provide starting points for the main topics in the Oasis stream; the room links also give access to the other sessions.

Four starting points in the Oasis stream:

Topic	Starting point in the recording
National AI strategies and the Blueprint	from 01:11:44
AI, work, education and skills	from 02:47:32
Introduction of the first scientific thematic brief	from 05:42:11
Building a shared scientific evidence base: Panel of Panels	from 06:07:11

The timestamps refer to elapsed video time and mark the beginning of each topic.

My initiative in the UN Portal: Robotics & AI 2050+

My Delphi study is now also listed as an initiative in the UN Digital Cooperation Portal: Robotics & AI 2050+. The direct entry in the MUDT/TUM profile describes a global, independent study of the future of robotics and artificial intelligence. It brings together expert assessments and covers six enabling technological capabilities and nine application domains.

The Portal maps the initiative to areas including international AI governance and inclusion in the digital economy. The study remains an independent research undertaking; the listing makes it discoverable in the international cooperation directory.

For me, these two tasks complement each other. The study aims to develop a systematic understanding of experts' views of the future. The one-page proposal aims to enable people outside the specialist community to contribute their own questions and expectations.

Part 2 · Personal Commentary: One Page That Gives People a Say

The following reflections express my own position. I raised the proposal at the event and want to put it up for discussion.

One proposal, several opportunities to raise it

I attended the UN Digital Cooperation Day in New York as an expert.

The visit included many interesting conversations with people who help shape the AI debate. I raised my one-page proposal in several settings: in conversations and in a contribution framed as a question to a panel.

My concern is a practical connection between research and the public. If we develop technologies that may change many people's everyday lives, those people need an understandable basis for forming a judgement. They also need to be able to raise their questions.

My proposal starts with something we can take responsibility for as researchers and developers: explaining what we do, what we know and what remains uncertain.

Dominik Bösl (left) and Gary Marcus in a selfie together in New York. — Guess who I bumped into in New York — and promptly roped into a selfie. Gary Marcus was a little caught off guard, but we had a lovely chat afterwards.

A scientific paper serves a different purpose from a public explanation

Scientific publications primarily address fellow specialists. They should make methods, results and conclusions open to scrutiny. That requires technical precision and the terminology of the field.

People outside that field need an understandable way in as well. They help fund research through their taxes. They encounter its applications in everyday life. Some benefit; others have to deal with consequences they did not choose.

That is why the proposal explicitly includes us as professors. We should be able to explain the problem we are working on, which expectations we consider justified and where we still have questions ourselves.

Writing one understandable page requires choices. What do people need to know to make sense of the project? Which terms need explaining? Which limitation must I avoid hiding behind an impressive demonstration?

Six questions for publicly funded technology

For me, the page should answer six questions.

Problem & people affectedWhat are we trying to solve – and for whom?
Technology & alternativesWhy this approach? What else could work?
Demonstrated capabilitiesWhat already works – and under which conditions?
Limits & what is unknownWhat can we not do, or do not yet know?
Risks & responsibilityWho decides on safeguards and when to stop?
Questions & participationHow can the people affected influence decisions?

1. What problem are we trying to solve, and who would benefit?

Start with the need. Who is meant to benefit from the project? What would an improvement look like? The explanation should also address who might face disadvantages or additional demands.

2. Why do we need this technology? What alternatives are available?

A technical possibility does not, by itself, justify its use. There may be a simpler way to solve the problem. A combination of technical and organisational changes may be needed. The public should be able to understand why a particular approach was chosen.

3. What demonstrably works already, and under what conditions was it tested?

A promise needs a basis. What has actually been tested? Under which conditions? How much of that can already be transferred to the intended application? Understanding the difference between an experiment and reliable operation helps people assess expectations.

4. What can we not do yet, and what do we not yet know?

An honest “We do not know yet” is a useful answer. It makes uncertainty visible. This includes technical limitations and open questions about the consequences of a possible deployment.

5. What risks do we see? Who decides on safeguards or stopping the project?

Identifying risks needs to affect the design. Who is responsible? What safeguards are planned? Who can intervene if assumptions no longer hold or problems arise? These responsibilities should be understandable too.

6. How can affected people ask questions and influence further development?

A contact address can be a starting point. What matters is what happens next: when can people contribute? Who answers their questions? How will they find out whether an objection changed a decision?

I also summarised the proposal and the six questions in my LinkedIn video post of 30 September 2026.

What this might look like in practice

Imagine a publicly funded research project developing a service robot for public buildings. Its purpose is to help visitors find their way. This is a hypothetical example to illustrate the proposal.

The public explanation could first identify the navigation problem being studied and the people the support is intended to help. It would then need to explain why a robot is being tested for this purpose. Better signs, an accessible building map or a staffed information point would be obvious alternatives to compare.

To explain its capabilities, the project would need to describe what works so far and under which conditions. Does the system recognise differently worded questions? Can it handle several languages? What happens when it does not understand a question? Which tasks will a person continue to handle?

If the project needs sensors or personal data, the explanation should describe why they are needed and which questions remain open. It should also make clear how people can reach their destination without using the robot.

Finally, the project needs feedback from everyday use: from visitors, employees and people whose needs may have been overlooked in the first design. Their suggestions should still be able to influence the design.

That turns a general announcement into a project people can discuss in practical terms.

Basic research needs room for what we do not know

For basic research, the honest answer to the question of practical benefit may be: we do not know yet. Advancing knowledge is a legitimate aim. I do not want every scientific question to have to justify itself prematurely with a supposedly certain business case or an immediately measurable everyday benefit.

Even then, it is possible to explain the question being studied, why it is scientifically interesting and what uncertainties remain.

My proposal does not require publishing trade secrets or security-sensitive details either. A public explanation must respect those boundaries. Anyone promising a benefit should, however, be able to explain the basis for that promise and what is still an expectation.

The page opens the conversation

One page alone does not create participation. I therefore want it to come with an accessible contact person and an accountable way of handling feedback.

What questions were asked? What objections were raised? Who decided how to respond? What changed, what was deliberately retained, and why?

The point is not to accept every suggestion. Participation becomes credible when decisions are explained and people can see what their contribution achieved.

This approach connects my thinking on Robotic & AI Governance with the practical approach in AI After 60: explain clearly, make independent checking possible and help people form an informed opinion.

The same question arises in social media regulation for young people: how can protection, education and self-determination work together? Support must fit the people and their circumstances. Understandable information and meaningful opportunities to have a say remain fundamental tasks.

A pilot measured by how well people understand

The objection “more bureaucracy” is justified. My proposal is therefore to integrate the page into existing project reports and update it at important milestones. I do not want an additional form that ends up circulating only between the project and its funding body.

I would start by piloting the requirement in one funding programme. For me, that includes testing the explanation with people outside the field. Can they describe the purpose in their own words after reading it? Can they tell what already works and what is still uncertain? Can they find the right route for asking a question?

It would be equally important to assess whether feedback arrives in time to influence further development. Those are the questions by which I would judge the pilot.

International scientific cooperation remains important. My assessment of the UN's scientific AI Panel considers that wider level. An understandable explanation of a particular project gives people access to decisions that affect them directly.

I would like us to start. Anyone receiving public funding should be able to give the public an understandable account of their work. One page is a manageable first step. Its value will depend on the conversation and participation we build around it.

Frequently Asked Questions About the UN Outcomes and the One-Page Proposal

About the meeting and its outcomes

What was the UN Digital Cooperation Day 2026?

The UN Digital Cooperation Day 2026 was an event organised by the UN Office for Digital and Emerging Technologies on 21 September in New York. It focused on science, policy and capacity-building for a global AI future. Governments, researchers, businesses and civil society came together to exchange perspectives. The official programme sets out the context and individual sessions.

What concrete outcomes did the UN Digital Cooperation Day produce?

Documented outcomes include the launch of the UN Blueprint for National AI Strategies, the scientific AI Panel's first thematic brief, and a new ODET-UNDRR partnership on AI and disaster risk reduction. Topics discussed also included work, open AI and cross-border sandboxes. The UN event report and the thematic brief's publication page document these different kinds of outcomes.

Were new binding global AI rules adopted there?

The official materials reviewed here document practical guidance, scientific assessment, cooperation and discussion. They do not record the adoption of a binding global AI law or of my one-page proposal. The precise document type and its status therefore matter. A scientific assessment informs policy decisions; my proposal for publicly funded technology remains clearly marked as personal commentary. Sources: UN event report, the Panel's mandate and publications.

How does the September thematic brief differ from the preliminary UN AI report?

The preliminary report of 1 July 2026 assesses AI's opportunities, risks and impacts more broadly. The first thematic brief, released on 21 September, focuses on AI agents, misalignment and risks of losing human control. It explores one question in greater depth rather than simply replacing the overall report. The Panel explains its publication sequence in its own FAQ. My earlier blog post about the overall report retains its historical context.

What does the listing of Robotics & AI 2050+ in the UN Portal mean?

The UN Digital Cooperation Portal lists the global, independent Delphi study “Robotics & AI 2050+” as an active initiative in the MUDT/TUM profile. The entry describes its research approach and maps it to objectives of digital cooperation. This makes the initiative discoverable in the cooperation directory. The listing alone does not establish particular findings or the completion of a new survey round. The details are available in the direct UN Portal entry.

Where can I watch the UN event recordings?

Official recordings are available for Oasis, Bamboo North and Bamboo South. In the Oasis stream, the timestamps above lead directly to national AI strategies, work and education, the scientific thematic brief, and the Panel of Panels. To hear individual statements in context, viewers can continue watching the full stream from those points.

About my proposal for technology funding

What is a one-pager for publicly funded technology?

I mean one understandable public page about a research or technology project. It explains the problem being investigated, why the approach was chosen, what demonstrably works, where the limits and risks lie, and how affected people can have a say. After reading it, people outside the field should be able to ask informed questions. The page complements scientific publications by providing an accessible way in for the public. I also explain the proposal in my own video post.

Who should the one-page requirement apply to, and what should the page cover?

My proposal addresses anyone receiving public funding to research or develop technology. That explicitly includes universities and us as professors. I propose six answers: the problem and the people affected; the technology and its alternatives; demonstrated capabilities; limitations and what is not known; risks and responsibility; and opportunities to ask questions and influence decisions. This gives the public an understandable basis for judgement without requiring every reader to complete a specialist degree first.

How can basic research be explained when its eventual benefit is still unknown?

Basic research can explain its scientific question, why it matters and what remains open. A later practical benefit may be unknown. That is precisely what the page should then say in understandable language. I do not want to force a supposedly certain everyday benefit that the research cannot yet demonstrate. An honest account of what is not known helps the public distinguish new knowledge, expectations and substantiated results.

Would this require publishing trade secrets or security-sensitive details?

My proposal calls for an understandable explanation within justified limits of confidentiality. Goals, expected benefits, demonstrated capabilities, open questions and responsibilities can often be described without disclosing protected technical details. Where information has to be withheld, the public explanation should make that boundary understandable. The aim remains to enable people to assess a project and ask questions about it; the page is not a complete technical disclosure.

How could the one-pager be piloted without creating more bureaucracy?

I would initially integrate it into existing reports in one funding programme and update it at important milestones. What matters for the pilot is whether people outside the field can subsequently explain the project, recognise its limitations and direct a question to the right person. It should also assess whether feedback reaches the development process in time. For me, success means understandable information and workable routes to participation.

When does understandable information become meaningful participation?

Participation begins when questions and objections can influence decisions that are still open. That requires accessible contacts, suitable opportunities for dialogue and accountable responses. People should be able to see what changed because of their feedback, or why a decision was retained. Both technically informed objections and questions from people without specialist knowledge need a place. My articles on AI After 60 and Robotic & AI Governance develop this thinking further.

AI After 60: Try It, Think Critically, Have Your Say

Cover image: AI-generated illustration with fictional people; not a photograph of the Augsburg event.

Can I still get to grips with artificial intelligence after 60? I encountered that question at the Seniorentag 60+ event at Augsburg's adult education centre, the Volkshochschule, on 8 November 2025. My answer is yes. Start with something that interests you. Ask someone to explain an application, try it and check the results. This report on the Augsburg event provides the context for my talk.

I wanted to give people encouragement and a basis for forming their own opinions. Experiencing where a technology helps and where it falls short makes it easier to say what you expect from it. That takes understandable explanations and opportunities to try things out.

For me, the question therefore leads directly to participation in society. How do people gain access to new technologies? How can they influence their development? And what options remain for those who want to keep doing important things offline?

Your year of birth does not determine your openness

I know older people who are more open to new technologies than many younger people. That is a personal observation, not a statistic about generations. It reminds me to see people as individuals, with their own interests and circumstances.

Someone who has had little opportunity to explore digital applications may need more time, a clear explanation or someone alongside them for a first attempt. Good teaching makes room for that. It takes questions seriously and lets people decide for themselves what they want to use a technology for.

Looking at other age groups helps clarify the aim. My article on social-media regulation for young people also considers the relationship between protection, digital education and self-determined participation. Support needs to suit a person's age and circumstances. For me, enabling people to reach their own judgments connects the two debates.

Why I use dog training to explain AI

In my talks, I show an optical illusion: a wave pattern turns into the face of a young woman when you narrow your eyes. The familiar “dog or muffin” comparison also draws attention to how easily similar visual patterns can confuse us. These examples help me make pattern recognition tangible.

To illustrate one aspect of learning, I also use an analogy with dog training. Imagine a dog; let's call him Bello. He is learning to sit in response to a cue. His attention is drawn to a hand holding a small treat. The hand moves slowly over his head. Bello follows it, sits down and receives the treat along with praise. Dogs Trust also describes the basic principle of this training method.

Through repetition, a connection develops between the cue, the behaviour and a positive consequence. The hand movement can gradually be phased out. Eventually, the word “Sit” is enough. Bello has learned an association without having to follow a human explanation of the entire training situation.

For AI, the analogy illustrates how feedback can influence learning. In reinforcement learning, particular outcomes are evaluated, and the system is adjusted during training to make more highly rated outcomes more likely. The reward is a numerical quantity. The analogy explains this aspect of learning. Other AI training methods work differently, and it does not describe everything a dog can understand either.

The essential point for me is this: A convincing result alone proves neither human understanding nor subjective experience. AI can generate new texts, help with creative tasks and solve certain reasoning problems. Research shows, for example, improvements on verifiable tasks such as mathematics and programming, as well as support for writing stories. See DeepSeek-R1 in Nature and Doshi and Hauser on AI and creative writing.

A friendly turn of phrase therefore does not establish that the system has feelings of its own. And a confident answer does not establish that it is correct.

Start with something you want to do

In my Augsburg talk, I discussed smartphone translation features and help with research and planning, among other examples. I find concrete tasks like these useful starting points:

Have a short message translated, then compare a translation back into the original language with what you wrote.
Collect ideas for a day trip. Then check opening hours and transport connections with the relevant providers.
Ask for an explanation of a term and compare it with a reliable source.

Choose something manageable whose result you can judge. An abstract discussion then becomes your own experience: What helps me? What would I like to work differently? Where do I encounter limits?

If getting started feels difficult, trying something together can help. Your questions should guide the process. Someone with experience can support those first steps.

The key question: Does that make sense?

“Does that make sense?” is a simple question I keep asking when working with AI.

Outputs that sound plausible but are false or fabricated are often called AI hallucinations. I also discuss this limitation in my article “Blumentopf der KI” — why Germany still has a chance in AI. What matters in everyday use is this: an answer's tone does not tell us whether the information is correct. A verifiable source is therefore more useful than confidence alone.

Does the answer fit what I know? Can I check its source? Is the information still current? What do I need to confirm before acting on it?

A well-written itinerary does not confirm an opening time or a train connection. A plausible explanation deserves a second look when important decisions depend on it. Let AI make a suggestion, then check what matters for your purpose.

Trying things out and remaining sceptical go together. First-hand experience helps you form a considered opinion: I want to use this application. This result convinces me. Here I need more information or a different approach.

Understandable explanations make participation possible

People should be able to help shape technology. That is why I call for an understandable explanation of publicly funded technology: Anyone receiving public funding to research or develop technology should explain to the public, on one page, what they are working on. I have also made this one-pager proposal in connection with the UN Digital Cooperation Day in New York.

I have also explained the proposal in a German-language post about understandable technology.

That page should address the questions people need answered to form an opinion:

What problem is being addressed, and who benefits?
Why was this technology chosen, and what alternatives are available?
What has demonstrably worked, and under what conditions?
What are the limits and unanswered questions?
What risks are involved, and who is responsible?
How can people affected by the technology ask questions and influence decisions?

An honest “We don't know yet” belongs on that page too.

The explanation opens the conversation. Participation also needs people who can be contacted and a clear account of what happens next: What changed because someone raised an objection? What stayed the same, and why? People must be able to contribute their wishes while decisions can still be influenced.

I address the international dimension of this task in my analysis of the United Nations' scientific AI panel. That article focuses on shared foundations of knowledge and how societies approach AI's opportunities, risks and unanswered questions. The call for an understandable one-page explanation is my own proposal.

For me, that is part of a participatory dialogue about technology and of Robotic & AI Governance. My analysis of Zuckerberg's AI vision leads to the same question: How do we connect personal opportunities with public responsibility?

Participation in society also needs offline options

I want to encourage people to try digital technologies. At the same time, I call for a right to participate in society through non-digital means. People who cannot or do not want to use a smartphone must retain access to essential information, services and opportunities.

If an appointment can only be booked through an app, or important information is available only through a QR code, there is a practical question to answer: How can someone access the service if they cannot use that route?

Digital services can open up access. Good design also takes account of personal conversations, understandable information and accessible offline options. In my view, someone's choice of device must not determine whether they can take part in society.

My message from Augsburg remains: Try the technology, check its results and form your own opinion. Make that opinion heard. Shaping our technological future concerns us at every age.

I had already addressed the topic in October 2024 in my own video on AI and digitalisation after 60.

Frequently asked questions about AI after 60

Can I start using AI after 60?

Yes. Choose a manageable task whose result you can judge for yourself. An understandable explanation and support suited to your needs can make getting started easier.

What are useful ways to try AI in everyday life?

A first attempt could be a short translation, an explanation or a collection of ideas for a day trip. Check the result; for plans, that includes checking opening hours and transport connections against the providers' information.

Can I trust an AI's answers?

A confident tone is not enough. Check important claims for plausibility, sources and currency. What matters for a decision is whether the relevant information is reliable.

What does the dog-training analogy explain about AI?

It illustrates one aspect of learning: connecting a cue, behaviour and feedback. It is not a complete account of every AI method. Good performance alone does not prove human understanding or subjective experience.

What does a participatory dialogue about technology mean?

People should receive understandable explanations of technology and be able to influence its design at an early stage. That includes a clear account of how their questions and objections affected the outcome.

Why do offline options matter as more services become digital?

Because participation in society should not depend on using a particular device. My political aim is to expand digital opportunities while preserving accessible non-digital routes.

Further reading, October 29, 2026: How curiosity about AI connects with professional competence, judgment and early-career training is explored in my article on AI and the future of work.

“Oh No, Not Again…” — Why Germany Still Has a Chance in AI

Historical commentary. First published in German on 24 September 2025. This English translation incorporates the corrections made to the German text on 4 October 2026 and retains the original time perspective. English translation published on 16 October 2026.

Artificial intelligence is advancing rapidly, and it can often seem as though only the major players in the United States and China have a say in where it goes next. Yet this is precisely where a particular opportunity opens up for Germany and Europe. Three observations show where our potential for efficient AI lies.

Why language models hallucinate

In its research on “Why Language Models Hallucinate”, OpenAI examines why language models produce plausible but incorrect answers. The paper on arXiv shows, among other things, how training and evaluation can reward guessing when a model is uncertain. It does not prove that hallucinations are fundamentally impossible to resolve in every AI system. A 2024 study in Nature investigates statistical methods for detecting a particular subset of these errors.

In practice, this means that even the best models available at the time will continue to generate incorrect facts that can be difficult to spot. Bigger models, faster processing and more data alone will not solve this structural problem. This is precisely where research into fundamentally new approaches is needed.

What are hallucinations in AI models?

“Hallucinations” are responses from AI systems that sound linguistically plausible but are factually incorrect or entirely invented. One example would be an AI answering a question about a historical figure with a quotation that person never said. This is a structural problem of today's language models rather than simply a software bug. They work by predicting text on the basis of probabilities, rather than possessing a genuine understanding of truth. However, this way of working does not establish a general proof that the problem is unsolvable. For example, a system can refrain from giving a specific answer when it is uncertain.

The business model of the major technology companies

Companies such as NVIDIA, Microsoft and Google are currently pursuing ever larger models. The NVIDIA CES 2025 keynote and the GTC 2025 event clearly show how their commercial success rests on selling computing power. More models mean more GPUs and more revenue.

As analyses of the CES 2025 announcements show, the focus is on ever more powerful data centres and larger AI infrastructure. Efficient, lean AI is not their current priority, and I do not expect that emphasis to change fundamentally over the next few years. For the major providers, encouraging resource consumption is economically attractive; reducing it is less so.

Europe's opportunity in fundamental research

This is where Europe can make its mark. The future of AI will not be decided solely by the race to build ever larger models. New models, technologies and approaches are needed. German and European researchers can excel at precisely this intersection.

The EU adopted €7.3 billion in funding for research and innovation across several fields under the Horizon Europe work programme for 2025. The DIGITAL work programme for 2025–2027 includes a further €1.3 billion for deploying strategic digital technologies, including AI, cloud and data, cybersecurity and digital skills.

Fundamental research into efficiency, new algorithms and approaches that use fewer resources presents a major opportunity. Far from having lost our chance, we have an opportunity to gain a foothold and deliberately pursue new research directions. The prize is still up for grabs, and Europe has the means to compete for it.

A different approach in practice: Apple's on-device AI

With its Apple Intelligence features, Apple is pursuing a strategy that differs fundamentally from its competitors'. It brings many AI features directly to users' devices through on-device AI, complemented by larger server models for Private Cloud Compute.

As technical analyses show, this substantially reduces the resources required in data centres, strengthens privacy and opens up new possibilities for using AI in everyday life. The approach illustrates that innovation need not always mean “bigger”. Sometimes progress lies in efficiency.

At the same time, the market for small language models, or SLMs, is developing rapidly, alongside edge AI hardware setting new benchmarks in 2025.

Conclusion

The AI revolution is well under way, and Europe has every opportunity to keep pace and lead in crucial areas. While the major corporations pursue scale, Europe can take a leading role in fundamental research, efficiency and innovation. The time to act is now. We have not lost our chance to win.

Related reading added on 16 October 2026: For a later perspective on Europe's role in AI and the responsibility of technology providers, see my analysis of Zuckerberg's vision and European AI governance.

Related reading added on 18 October 2026: AI After 60 explains how critical checking can go together with an encouraging start: through everyday experiments, the dog-training analogy and the question “Does that make sense?”.

Zuckerberg's "The Future is for Everyone": Who Bears the Consequences?

Editorial status: 6 October 2026. An analysis of Mark Zuckerberg's essay "The Future is for Everyone" of 10 August 2026 – read through the framework of the 4 Robotic Revolutions, the encyclical Magnifica Humanitas, and a European approach to Robotic & AI Governance.

On 10 August 2026 Mark Zuckerberg published a manifesto at meta.com/thefutureisforeveryone. The text, close to 6,500 words, is one of the most direct statements a Big Tech CEO has offered on superintelligence in years. Zuckerberg writes that Meta is building "personal superintelligence" - an AI available to every individual on a persistent basis as tutor, advisor, business tool, and conversational partner. He argues against a single benevolent superintelligence, against doom rhetoric, against concentration of compute in few hands. He argues for distribution, open source, community compacts, and a US-led model in competition with China. The essay closes with the line "Superintelligence holds the promise of giving everyone that power, and building a positive future for everyone" - signed "- Mark" and dated August 10, 2026.

The text deserves careful reading. It also deserves careful pushback. My thesis for this post: Zuckerberg's essay asks the right question ("Who owns the future?") but delivers a structurally truncated answer. Reading the text as a blueprint for the coming decade of AI and robotics hands governance responsibility to a single company, a burden Meta neither can carry nor should carry. What the text is missing is the institutional substructure that Europe - and specifically Germany - could offer if we stopped making ourselves intellectually small.

What Zuckerberg actually says

Any critique deserves a fair rendering. Zuckerberg opens with an observation that is correct: the debate about superintelligence has so far been either eschatological (doom, existential risk, alignment failure) or utopian (curing cancer, solving physics). Both frames ignore, in his view, how the technology will actually settle into the everyday life of billions of people. His answer is an empowerment narrative. Every human being gets their own AI, one that supports them in work, education, social life, dealing with bureaucracy, and personal organisation. Meta is pursuing this work through its "Meta Superintelligence Labs", investing heavily in data centres, for example in Louisiana, pledging water-positive operations by 2030 (200 percent in water-stressed regions), and signing "Community Compacts" with the municipalities that host the compute. The example named in the text: Richland Parish, Louisiana, with teacher bonuses of USD 50,000 in 2026 from increased tax revenue, according to Zuckerberg.

Zuckerberg makes three philosophical moves. First: empowerment over automation. AI should not replace work; it should extend the capabilities of the individual (he calls this "invention" instead of "automation"). Second: distribution over concentration. He argues explicitly against a "single benevolent superintelligence" held in one place, even if its values were genuinely benevolent. Third: open source as a safety principle. When many actors have access to powerful models, so the reasoning, a "balance of power" emerges between AI systems and between humans and AI.

LeCun announced in November 2025 that he would leave Meta at the end of that year. He also envisaged a partnership between Meta and his new AMI company. His departure alone therefore does not establish a break with Zuckerberg’s position. LeCun’s own announcement.

Zuckerberg identifies recursive self-improvement (RSI) as a risk: systems could improve themselves and become harder to control. He proposes providing more compute for people than for such systems, involving Meta’s independent corporate directors in model-release decisions, and sharing intermediate training checkpoints with the US government. For me, the question remains how these internal measures would work alongside independent public oversight. Aligning AI more closely with individual users does not resolve the wider question of responsibility.

The philosophical question: individual or common good?

Zuckerberg's text is one of the sharpest articulations of a strictly individualistic understanding of AI I have read in years. The empowerment framework sits so deep in the essay that on a first reading it feels obvious. It is not.

The European philosophical tradition asks a different question. It does not begin with "what does AI do to me?" but with "what does AI do to us?". Immanuel Kant formulated this in the Categorical Imperative as the test of whether a maxim of action could be conceived as a general law. Applied here: can the maxim that every human being possesses a private superintelligence be a general law? And if so, what follows for the design of institutions, education, division of labour, and public space? In my view, the essay does not develop these collective preconditions sufficiently. He takes the individualistic model as given and derives institutions from it, rather than the other way around.

AI ethics in my lectures at the Hochschule der Bayerischen Wirtschaft (HDBW) raised questions also addressed by Pope Leo XIV's encyclical Magnifica Humanitas. I analysed it in an earlier post. The core idea: the human person has a value that is not exhausted by utility functions. Technology is not a neutral instrument. It is institutional action that defines chains of accountability and either builds them or breaks them. When an AI sits in every person's toolbox, the shift is not only in individual productivity. It is in the position of the person in the polity. She becomes potentially more autonomous – and simultaneously more dependent on the provider that controls the technical foundation of that autonomy. In my view, the essay gives this ambivalence too little attention.

My philosophical critique concerns the power relations behind the promise of empowerment. AI can give people new capabilities while making them more dependent on their tools’ providers. For me, a central lesson from the development of social platforms is that personal empowerment does not itself guarantee democratic control over infrastructure, data or access. Anyone promising superintelligence should therefore explain how individual freedom and public responsibility fit together.

The ethical dimension: who is liable for the consequences?

In a series of academic papers - including "Roboethics and Robotic Governance" (Bösl & Bode, 2017, in the Springer volume "ROBOT 2017: Third Iberian Robotics Conference", DOI 10.1007/978-3-319-70833-1_12) - I have developed the concept of Robotic & AI Governance. The core: from a governance perspective, autonomous systems (robots as much as AI models) cannot be understood solely as conventional tools. This does not mean they are exempt from product liability. Nor are they legal persons in the sense of civil law. They are institutional intermediary entities whose actions emerge from a chain of design, training, deployment, and use decisions. Governance is the ordering framework that makes this chain reconstructable and legally addressable.

Zuckerberg wants Meta’s independent corporate directors to oversee model releases. This is distinct from the Oversight Board for content moderation. My critique concerns a different question: how are internal decisions complemented by public rules, independent scrutiny and traceable responsibility? Zuckerberg also acknowledges the need for government policy, but leaves the European arrangements open. Users can configure their AI, yet do so within a technical framework shaped by the provider. Personal choices therefore do not replace public accountability.

What is missing is the recognition that governance cannot be self-governance. For any serious regulation of high-risk products - pharma, aviation, nuclear energy, chemistry - the separation between provider, independent testing body, and state supervision is constitutive. The EU AI Act, in force since 1 August 2024, with phased application dates, provides for conformity assessment and public supervision; an independent testing body is not required in every case. Zuckerberg does not mention the EU AI Act in his essay. He does not mention Europe at all. I read this as a deliberate omission and as a strategic self-disclosure: in the world order Zuckerberg imagines, Europe is not an actor. It is a market.

A second ethical question concerns alignment: what should guide an AI? Zuckerberg emphasises individuals and their goals, a plausible promise of personal support. But if individual preferences become the sole criterion, problematic or shortsighted wishes may also be amplified. In my view, this does not settle the question of shared rules and the common good. Reconciling personal empowerment with other people’s rights requires political and public debate.

The socio-political dimension: distribution as a power strategy

Zuckerberg’s argument for open models and distribution is also geopolitical. The essay contrasts the US and China and addresses export controls and American leadership. For me, this raises an institutional question: how do states and multilateral institutions retain the ability to shape developments when powerful AI becomes widely available? Distribution can empower people. It also requires rules whose enforcement does not depend solely on providers’ goodwill.

Distributing applications does not automatically distribute control over their foundations. Those who train models, operate infrastructure and set access conditions retain considerable influence. For me, this is the tension in Zuckerberg’s balance-of-power argument: Meta can make applications available to many people while controlling central decisions about them. Whether that produces a workable balance depends on the specific rules governing access, scrutiny and responsibility.

The first responses show different emphases. The Register and 404 Media commented sharply on the essay on 10 August 2026. The ETC Journal recognised the importance of the question while challenging the answers. These three voices are examples of the early reception, not a systematic survey of the field.

For European policy, the task is clear: formulate its own institutional answers. The EU AI Act is a comprehensive European legal framework. Its implementation will show where it needs clarification and how effectively supervision works. An internal company body can complement that public task, but cannot take it over.

The socio-economic dimension: community compacts and labour displacement

Zuckerberg's essay describes "Community Compacts" - agreements between Meta and the municipalities that host the data centres. The example: Richland Parish, Louisiana. Zuckerberg reports teacher bonuses of USD 50,000 in 2026 from increased tax revenue, supports school infrastructure, and commits to water-positive operations by 2030 (200 percent in water-stressed regions). This sounds generous. It is also a striking shift of the social contract. According to the essay, the bonus is enabled by additional local tax revenue. My institutional question remains: how much room does a municipality have to shape its future when its development depends so heavily on one company? Additional local revenue does not settle how benefits and burdens are distributed within a region and beyond it.

Such a site agreement cannot simply be transferred to Europe unchanged. Additional local revenue and corporate commitments need to work alongside existing public responsibilities, education systems and welfare arrangements. If Germany wants to attract data centres, the question of regional benefit should therefore go beyond a single bonus: vocational education, basic research, social partnership and regional value creation should, in my view, be part of the negotiations.

In my view, the essay does not sufficiently develop the question of work. According to the International Federation of Robotics, around 4.66 million industrial robots were operating worldwide in 2024, nine percent more than in the previous year. World Robotics 2025, Executive Summary. That shows the growing importance of industrial automation. The figure alone does not tell us how many jobs are created, transformed or lost. We need to investigate those effects rather than resolve them rhetorically through a contrast between invention and automation.

In an earlier essay on the 4 Robotic Revolutions I described how robotic industrialisation unfolds in four phases: protected industrial robots, sensitive and collaborative robotics, mobile systems, and cognitive robotics with increasing autonomy. The framework helps distinguish different changes in work and organisation. Anyone assessing generative AI should therefore also consider the experience of industrial automation.

The legal dimension: EU AI Act and provider responsibility

The EU AI Act has been in force since 1 August 2024, with phased application. Provider obligations for general-purpose AI models are set out primarily in Article 53; Article 55 adds requirements for models with systemic risk. Subject to statutory exceptions, these include technical documentation, information for downstream system providers, a copyright compliance policy and a public summary of training content, with additional safety evaluations for models with systemic risk. For Meta’s models, GPAI obligations need to be assessed for the specific version and its placement on the EU market. Transition periods matter: models placed on the market before 2 August 2025 generally have until 2 August 2027 to comply. MSL had already presented Muse Spark in April 2026. The essay does not replace an assessment of European provider obligations.

The legal question is not theoretical. If Meta offers "personal superintelligence" in Europe, these systems fall under the EU AI Act. Additional GPAI provider obligations depend on the actor's role in placing the model on the market, not simply on open weights. Article 53 provides limited exceptions for certain models released under free and open-source licences; those exceptions do not cover models with systemic risk. Whether the high-risk regime applies depends on the specific intended purpose under Article 6 and Annex III, not simply on the sector. Under the amended timetable, the main obligations for Annex III systems apply from 2 December 2027. This timetable was amended by the Digital Omnibus on AI in July 2026. For me, the crucial point is that personal empowerment in a European rollout must be connected with verifiable provider and deployer responsibility.

Copyright cannot be reduced to a blanket admission by Meta. In Kadrey v. Meta, the US court found the training use at issue to be fair use against the 13 plaintiff authors on the specific evidentiary record on 25 June 2025. That was not a general authorisation of all AI training. US court decision. In France, SNE, SGDL and SNAC made allegations against Meta and announced a claim in March 2025; those allegations should be distinguished from judicial findings. The organisations’ announcement. For European governance, the origin of training material and traceable rights clearance remain central questions.

The 4 Robotic Revolutions as an analytical frame

Together with Bernd Liepert, I proposed the "4 Robotic Revolutions" model at IEEE/RSJ IROS in Daejeon in 2016. The first revolution describes the protected industrial robot. The second concerns sensitive and collaborative robotics and cooperation with people. The third is mobility: robots leave their fixed workplace. The fourth concerns cognitive robotics and increasing autonomy. The phases overlap; they are not a rigid chronology. Humanoid systems and generative AI are developments that can be analysed through this framework, rather than additional phases of their own.

Zuckerberg’s essay considers AI primarily through digital models and personal assistants. In my view, the connection with physical robotics receives too little attention. Over the coming decade, I expect generative AI, humanoid systems, autonomous mobility and industrial automation to become more closely connected. That calls for a careful examination of companies’ technical capabilities and investments. Unitree was preparing an IPO on Shanghai’s STAR Market in 2026. Its March application prospectus envisaged around RMB 4.2 billion for investment projects. This was a financing plan, not evidence of a capital inflow already completed at that time.

China’s weight in industrial robotics is already clear: according to the IFR, China accounted for around 54 percent of new industrial robot installations worldwide in 2024. IFR data for 2024. That makes it the largest installation market in this category. The figure alone does not establish leadership across all areas of physical AI or humanoid robotics. An analysis of US-China competition should therefore distinguish digital models, industrial deployment and new robot forms.

The fourth robotic revolution will demand a European response that does not copy Meta. It will need the coupling of generative AI, cobot deployment in the Mittelstand, humanoid robotics in structured environments (care, skilled trades, logistics), and a Robotic & AI Governance infrastructure that builds trust. Germany has the preconditions: social partnership, dual vocational education, the Fraunhofer system, academic research, TÜV, DIN, statutory accident insurance institutions (Berufsgenossenschaften). What is missing is the self-authorisation to position these institutions as robotic governance infrastructure.

Magnifica Humanitas and the European counter-narrative

In my earlier blog post on Magnifica Humanitas, I analysed Pope Leo XIV's encyclical. I also read it as a touchstone for the values that could carry a European AI policy. The core claim of my assessment: the human person is more than her preferences. She is a person, embedded in social relationships, historically grown institutions, culturally shaped educational trajectories, and legal orders. AI governance cannot retreat to the optimisation of individual preferences without damaging this personal dimension.

Zuckerberg's essay is, in this reading, the counter-position. He takes preferences as given, AI as an instrument for preference optimisation, and institutions as external obstacles or flexible negotiation partners (see: Community Compacts). The Magnifica Humanitas position argues the reverse: preferences are shaped, education is the institutional site of that shaping, and AI governance must protect the institutions of education before delivering the tools.

Concretely, for European education policy this means: schools do not need a "personal superintelligence" per pupil. They need a critical engagement with generative AI, teachers competent to shape its use, and examination formats that make the difference between tool use and independent thinking transparent. This connection between technology ethics, application and responsibility has been part of my teaching for many years. The point is not to ban AI. The point is to preserve the institutional setting in which learning takes place as a social practice.

For health care the position means: AI support in diagnostics and treatment planning, yes. But under the condition that the doctor-patient relationship remains protected as a personal relationship. The sick person does not need a "personal superintelligence" that gives her PhD-level advice. She needs a doctor who integrates AI support competently. The difference is not semantic. It is institutional.

For public administration, my position is that AI can support caseworkers, but responsibility and the competent authority must remain traceable. The law does not require a human decision-maker for every decision: section 35a of Germany’s Administrative Procedure Act permits fully automated administrative acts under specified conditions. For me, the decisive question is how citizens can understand decisions, challenge them and obtain effective review. Enthusiasm for efficiency must not obscure that question.

Where Zuckerberg is right

I share several underlying ideas in the essay. First, debate about existential risks should not displace attention to AI’s everyday effects. Second, the concentration of great technical power deserves scrutiny even when justified by good intentions. Third, open models can facilitate independent scrutiny and research. What matters is what is actually accessible: code, weights, documentation or training data. Open weights alone neither enable full reproduction of training nor automatically guarantee safety. Fourth, the municipalities hosting data centres should help shape their development and share in the benefits. Community Compacts offer one approach, whose design should be assessed against the relevant public framework.

One strength of the essay is that Zuckerberg puts his position forward under his own name. He identifies risks, investment plans and commitments, giving us specific statements against which the strategy can be assessed. A public promise is not evidence of delivery. That is why we should read the essay as an invitation to scrutiny and measure later decisions against it.

My agreement ends where the text brackets out the institutional preconditions of a superintelligence distribution. It ends where internal company oversight appears to be a sufficient answer to questions of public responsibility. It ends where Europe is not mentioned. It ends where automation is renamed empowerment without settling accounts with the past fifteen years of Facebook history.

What Europe and Germany should concretely do

First: take the EU AI Act seriously and sharpen its implementation. The provider obligations for general-purpose AI have to be operationalised. The European Commission's AI Office needs resources, technical expertise, and the political backing to act against Meta, OpenAI, Google, and Anthropic when they refuse compliance.

Second: treat German dual vocational education and academic training in robotics and AI as strategic infrastructure. Not as a cost item. In my view, deployment skills deserve as much attention as technological development. The HDBW, the TUM, the Fraunhofer institutes, the vocational academies, and dual vocational training are the institutions that can build robotic deployment competence. They need resources and political recognition.

Third: develop Community Compacts adapted to Europe. In my view, site negotiations should include structural investment in regional value chains, cobot pilots for the Mittelstand, and links between data centre development and vocational education. Bavaria could take a lead. What matters is that states, municipalities, companies and education institutions agree shared goals and make delivery verifiable.

Fourth: position Robotic & AI Governance as a German export good. We have the institutions (TÜV, DIN, VDE, Berufsgenossenschaften). We have the scientific base (Fraunhofer, DFKI, TUM, KIT, RWTH). We have the legal tradition (administrative law, product liability, data protection). We have the European legal bracket (EU AI Act). What is missing is the natural sense that this governance competence is a strategic resource, not an inheritance to be administered. The world market will demand governance standards for autonomous systems in the coming years. Those standards can be German-shaped, if Germany offers them in time.

Fifth: free the debate from its individualistic framing. Superintelligence is not a question of individual empowerment. It is a question of the institutions that decide how empowerment is distributed, controlled, held to account, and shaped. The European answer to Zuckerberg is not a better empowerment promise. It is an institutional self-authorisation with the courage to formulate AI governance as an original civilisational contribution.

What this means for my teaching and practice

Technology ethics has been part of my teaching for many years. At HDBW, I have addressed these questions over numerous semesters. At TUM, I cover them in several of my courses across all MBA programmes and in a focus area. From October 2026, I will also teach a course at MUDT that brings together technology ethics, governance, sustainability and responsible innovation.

I have also addressed current developments and this debate in my keynotes at ABB, DEVELOP and People and Robots in Sicily, with a different emphasis in each talk.

AI is global politics – and Germany needs to take a clearer position

AI is now an issue of global politics. The discussions at the United Nations, including the Global Dialogue on AI Governance, make that clear. They address questions that go far beyond technical performance: Who gets access? Who sets the rules? Who bears responsibility? And how can people influence developments that change their everyday lives? UN account of the Global Dialogue.

These questions will stay with us for years to come. I have yet to see a clear, publicly visible German position. Germany should contribute its experience in technology ethics, industrial applications and public participation far more decisively to this international debate.

For me, AI governance therefore needs a scientific foundation and participatory technology dialogue. I will explore what the UN report can contribute and what this means for Germany in a forthcoming article on that report.

My recommendation to everyone taking a position in this debate: read Zuckerberg's essay. Literally, not in summary. Note the omissions. Note the rhetoric. Then formulate the European position - not as a reaction, but as a stand-alone civilisational statement. We have the institutions. We have the intellectual tradition. We have the legal bracket. What we lack is the natural sense that this equipment is a strategic resource and not an inheritance we merely manage.

The future is not "for everyone". It belongs to those who refuse to let anyone else answer the question about the institutions of their future for them.

FAQ on Zuckerberg's "The Future is for Everyone"

What is Zuckerberg's essay „The Future is for Everyone"?

“The Future is for Everyone” is an essay published by Mark Zuckerberg on 10 August 2026. The roughly 6,500-word text describes his vision of personal superintelligence: AI supporting people in everyday life. He argues for broad availability, open models and American leadership in competition with China. The text also describes the work of Meta Superintelligence Labs, data centre investments and Community Compacts with host municipalities.

What is the core critique of Zuckerberg's position?

My core critique is that personal empowerment needs an institutional framework. Zuckerberg addresses government policy and the role of Meta’s corporate directors; in my view, this does not sufficiently answer questions of public responsibility. The essay does not mention the EU AI Act. If Meta distributes applications while controlling their technical foundations, questions of independent scrutiny, access and accountability remain.

What does Zuckerberg mean by „personal superintelligence"?

“Personal superintelligence” describes Zuckerberg’s vision of a personal AI assistant supporting people in learning, working and planning, oriented towards their everyday lives. That vision should be distinguished from an available product with established technical properties. The models behind a specific application, and the extent to which their weights, code or training foundations are accessible, need to be checked separately.

How has the field responded to Zuckerberg's essay?

FAQ on Robotic & AI Governance and the EU AI Act

What is Robotic & AI Governance?

Robotic & AI Governance is the framework that makes the development, training, deployment and use of autonomous systems reconstructable and legally addressable. Its academic background includes "Technology Governance" by Dominik Boesl and Martina Bode (IEEE EmergiTech, 2016) and "Roboethics and Robotic Governance – A Literature Review and Research Agenda" in the Springer volume "ROBOT 2017: Third Iberian Robotics Conference" (2017). The core: from a governance perspective, autonomous systems are more than conventional tools; this does not exempt them from product liability. Nor are they legal persons in civil law. From a governance perspective, they are institutional intermediary entities whose actions emerge from a chain of design, training and use decisions. Governance makes that chain addressable.

How does Zuckerberg's proposal relate to the EU AI Act?

The essay does not mention the EU AI Act, which has been in force since 1 August 2024 with phased application. GPAI provider obligations depend on the specific model version, placement on the EU market and the provider’s role. Article 53 covers documentation, information for downstream providers, a copyright policy and a public summary of training content, subject to limited exceptions. Article 55 adds obligations for models with systemic risk. Models placed on the market before 2 August 2025 generally have a transition period until 2 August 2027. Muse Spark was already presented as an MSL model in April 2026. A model name alone does not establish a violation.

Why is Zuckerberg's „independent board" problematic?

Zuckerberg wants Meta’s independent corporate directors to take part in model-release decisions. This is distinct from the Oversight Board for content moderation. My critique is that internal company oversight does not replace public rules, independent supervision or traceable responsibility. How these levels should work together needs to be clarified for European deployment.

How does Zuckerberg redefine alignment?

Zuckerberg emphasises aligning AI with individuals and their goals. That can improve personal support. But if individual preferences become the sole criterion, problematic or shortsighted wishes may also be amplified. In my view, the question remains how personal empowerment can be reconciled with shared rules, other people’s rights and the common good.

FAQ on the 4 Robotic Revolutions and Magnifica Humanitas

What are the 4 Robotic Revolutions?

The 4 Robotic Revolutions are a phase model presented by Dominik Boesl and Bernd Liepert at IEEE/RSJ IROS in 2016. First: protected industrial robotics. Second: sensitive and collaborative robotics and cooperation with people. Third: mobile robotics that leaves the fixed workplace. Fourth: cognitive robotics and increasing autonomy. The phases overlap rather than forming a rigid timeline. The model helps to analyse current developments, including the connection between generative AI and physical robotics.

What does Magnifica Humanitas mean?

Magnifica Humanitas is an encyclical by Pope Leo XIV, dated 15 May 2026. I analysed it in an earlier blog post. My assessment connects it with a personalist perspective on European AI governance: the human person is more than her preferences. She is embedded in social relationships, historically grown institutions, culturally shaped educational trajectories, and legal orders. AI governance cannot retreat to optimising individual preferences without damaging this personal dimension. I see this as a counter-position to Zuckerberg's individualistic empowerment frame.

Why does Zuckerberg ignore physical robotics?

The essay considers AI primarily through digital models and personal assistants. This does not establish an intention to exclude physical robotics. My critique is analytical: the connection between generative AI, mobility and industrial automation deserves more attention. According to the IFR, China accounted for around 54 percent of new industrial robot installations worldwide in 2024. That establishes its weight in this category, not general leadership in humanoid robotics.

How do generative AI and physical robotics interlock in the fourth revolution?

The fourth robotic revolution is not shaped by generative AI alone. It combines generative AI with humanoid robotics, autonomous mobility, and industrial automation. The coming years will show how cobots in the Mittelstand, humanoid systems in structured environments (care, skilled trades, logistics), and generative AI assistants work together. Germany has the institutions (social partnership, dual vocational education, Fraunhofer, TÜV, DIN) to structure this transition - if the self-authorisation to do so is in place.

FAQ on European AI policy and German institutions

What should Europe do in response to Zuckerberg's essay?

Five concrete steps. First: take the EU AI Act seriously and sharpen its implementation. The European Commission's AI Office needs resources and political backing. Second: treat dual vocational education and academic training in robotics and AI as strategic infrastructure. Third: develop community compacts in a European mode - regional value chains, cobot pilots, vocational education investment tied to data centre siting. Fourth: position Robotic & AI Governance as a German export good. Fifth: free the debate from its individualist framing.

Why is the American Community Compacts approach not directly transferable to Europe?

Zuckerberg describes agreements with host municipalities and reports teacher bonuses funded by additional tax revenue in Richland Parish. Such local arrangements cannot simply be transferred to Europe unchanged. Corporate commitments need to work alongside existing public responsibilities, education systems and welfare arrangements. In my view, regional value creation, vocational education and verifiable shared goals should be part of the negotiations, beyond individual bonuses.

What role does Germany play in European AI governance?

Germany has the institutions (TÜV, DIN, VDE, Berufsgenossenschaften), the scientific base (Fraunhofer, DFKI, TUM, KIT, RWTH), the legal tradition (administrative law, product liability, data protection), and the European legal bracket (EU AI Act). What is missing is the natural sense that this governance competence is a strategic resource. The world market will demand governance standards for autonomous systems in the coming years. Those standards can be German-shaped - if Germany offers them in time.

What does the UN debate on AI governance mean for Germany?

AI governance is an issue of global politics. The UN debate addresses access, responsibility and public participation. In my view, Germany should take a much clearer position and contribute its experience in technology ethics, industrial applications and participation. These questions will remain relevant for years to come. The UN report provides an opportunity to discuss the scientific foundations and political implications separately.

Sources

Zuckerberg, Mark (2026): “The Future is for Everyone.” Meta, 10 August 2026. Original essay.
Claburn, Thomas (2026): “The future is for billionaires – the rest of us will get open weight AI models, maybe.” The Register, 10 August 2026. Original commentary.
Koebler, Jason (2026): Commentary on the Zuckerberg essay. 404 Media, 10 August 2026. Original article (headline and date publicly accessible).
ETC Journal (2026): “Zuckerberg Got the Question Right, but Answers Are Up for Grabs.” 10 August 2026. Original article.
Bösl, Dominik B. O. & Bode, Martina (2017): "Roboethics and Robotic Governance - A Literature Review and Research Agenda." In: Ollero, A., Sanfeliu, A., Montano, L., Lau, N. & Cardeira, C. (eds.): ROBOT 2017: Third Iberian Robotics Conference. Advances in Intelligent Systems and Computing 693, Springer, Cham. DOI: 10.1007/978-3-319-70833-1_12
Bösl, Dominik B. O. & Liepert, Bernd (2016): "4 Robotic Revolutions - Proposing a Holistic Phase Model Describing Future Disruptions in the Evolution of Robotics and Automation and the Rise of a New Generation 'R' of Robotic Natives." In: 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Daejeon, Korea, pp. 1262-1267. DOI: 10.1109/IROS.2016.7759209
European Union (2024): Regulation (EU) 2024/1689 of the European Parliament and of the Council - "EU AI Act". In force since 1 August 2024, with phased application dates.
Verordnung / Regulation (EU) 2026/1744: Digital Omnibus on AI, 8 July 2026; Official Journal 24 July 2026.
International Federation of Robotics (2025): World Robotics 2025. Executive Summary.
United Nations (2026): From AI to killer robots: UN chief issues urgent governance call, 7 July 2026.
LeCun, Yann (2025): Own announcement of departure from Meta, November 2025.
Meta (2025/2026): Llama 4 (5 April 2025); Muse Spark / Meta Superintelligence Labs (8 April 2026).
European Commission: Guidelines on obligations for General-Purpose AI providers.
US District Court, Northern District of California: Kadrey v. Meta, order of 25 June 2025.
SNE / SGDL / SNAC (2025): Announcement of claim against Meta, March 2025.
Unitree (2026): STAR Market application prospectus, March 2026.
Oversight Board: Our work (content moderation).
§ 35a VwVfG: Fully automated administrative acts.
Bösl, Dominik (2026): "Magnifica Humanitas: The Encyclical and AI Governance - An Assessment." boesl.org, 16 June 2026. boesl.org
Bösl, Dominik (2026): "Four Robotic Revolutions: 10 Years On." boesl.org, 25 July 2026. boesl.org
Bösl, Dominik (2026): "EU AI Act August 2026: What Really Takes Effect." boesl.org, 23 June 2026. boesl.org

About the Author

Since 1 October 2026, Prof. Dominik Boesl has been Professor and Chair of Applied Robotics and Intelligent Automation (ARIA) and Founding Director of the Center for Robotics at the Munich University of Digital Technologies & Applied Sciences (MUDT). Until September 2026, he was a professor at the Hochschule der Bayerischen Wirtschaft (HDBW). He is also an Adjunct Professor at the Technical University of Munich (TUM) and CEO of Innomancer GmbH. His work connects applied robotics with Robotic & AI Governance and the responsible deployment of intelligent machines.

Selected publications:

"4 Robotic Revolutions – Proposing a Holistic Phase Model …" (IEEE/RSJ IROS, 2016; DOI 10.1109/IROS.2016.7759209)
"Roboethics and Robotic Governance – A Literature Review and Research Agenda" (Springer, 2017; DOI 10.1007/978-3-319-70833-1_12)
Bösl, D. B. O. (2019): Predicting the Future of Robotics and Making It Matter for Industry [Industry Activities]. IEEE Robotics & Automation Magazine, 26(4), 128–130.
Mai, V. et al., including Bösl, D. B. O. (2022): The Role of Robotics in Achieving the United Nations Sustainable Development Goals – The Experts’ Meeting at the 2021 IEEE/RSJ IROS Workshop [Industry Activities]. IEEE Robotics & Automation Magazine, 29(1). DOI: 10.1109/MRA.2022.3143409.
Boesl, D. B. O.; Bode, M.; Liepert, B.; Greisel, S. (2017): Structured Megatrends Research as Foundation for Future-Oriented Research-Planning and R&D Roadmapping in Robotics. CTCEEC, Mysore. DOI: 10.1109/CTCEEC.2017.8455126.
Further publications at boesl.org/publications

Contact: dominik@boesl.org
LinkedIn: linkedin.com/in/boesl
Website: boesl.org

UN Independent Scientific Panel AI - Report Analysis 2026

Update, 4 October 2026: The first Global Dialogue on AI Governance took place in Geneva on 6–7 July 2026. On 21 September, the Panel published its first thematic brief on AI agents, misalignment and the risk of losing human control, as an advance unedited version. The second Dialogue is scheduled for 3–4 May 2027 in New York. The following essay, published on 2 July 2026, retains its original perspective.

On 1 July 2026, shortly after eleven in the morning New York time, the Independent International Scientific Panel on Artificial Intelligence released its first Preliminary Report. Forty scientists from all five UN regions, a three-year mandate, and one single task: to build the first globally independent scientific foundation on AI. Not a political paper, not a regulatory proposal, not a lobby text. A reference document that the United Nations will place before its 193 member states on the eve of the first Global Dialogue on AI Governance in Geneva on 6 and 7 July.

I have read the report, the personal message from the two Co-Chairs Yoshua Bengio and Maria Ressa that accompanies it, and the remarks of UN Secretary-General António Guterres at the launch. And I have tried to connect the material to what many of us have been working on for years: the question of a workable order for autonomous machines and for artificial intelligence, for Robotic & AI Governance. The report is a milestone. Not because it produces new facts, but because it puts existing facts, for the first time, into a common, scientifically independent text that all governments of the world will have to reckon with. And in a language that is politically no longer easy to brush aside.

What the report is - and what it is not

The panel was established by UN General Assembly Resolution A/RES/79/325 in August 2025, building on the Global Digital Compact of 2024 and the recommendations of the Secretary-General's High-Level Advisory Body on AI. It is the first global, standing scientific body devoted exclusively to AI. Not an IPCC clone, but with the same structural ambition: to document what science knows and what it does not know, without issuing prescriptions. The 40 members were selected from 2,600 applications across more than 140 countries. They serve in their personal capacity, for three years, free from instructions by governments, companies, or institutions. The Co-Chairs are the Canadian deep-learning pioneer Yoshua Bengio and the Filipina Nobel Peace laureate and journalist Maria Ressa. Two biographies that could hardly be more different. And precisely for that reason, their arriving at the same diagnosis carries more weight than either could alone.

The report runs to just under 60 pages of main text, organised in seven domains: AI science and development trajectories; societal applications (science, health, education, agriculture); economic implications; security and environmental impact; human rights, information and democracy; cultural and individual flourishing including child safety; and management, governance and reliability. It is explicitly preliminary - a first snapshot, with thematic addenda to follow before a comprehensive first full report in May 2027. And it is a scientific consensus document, not an opinion piece.

That is the first point I want to make clearly: the report is not a warning from activists. It is the strongest globally coordinated scientific assessment of AI we currently have. To ignore it is no longer to ignore an opinion. It is to ignore the state of international science.

Executive summary in my own words

If you have little time, these are the eight points to know. They distil what the report supports across 60 pages.

AI capabilities are growing faster than our ability to measure or govern them. On the Humanity's Last Exam benchmark, deliberately designed to be difficult, the best systems climbed from 8 to 45 percent within 16 months. On FrontierMath, from 19 to 88 percent in a little over a year. At the 2025 Mathematical Olympiad, several systems reached gold-medal level - much earlier than even domain experts had expected.
Frontier models are trained by a handful of actors. The United States holds around 75 percent of the compute in the world's top 500 AI supercomputers, China around 15 percent. Almost all leading general-purpose models originate in two countries. A small number of firms control critical inputs of the chip supply chain.
AI access and AI impact are extremely uneven, geographically and linguistically. More than one billion people use AI chatbots weekly, but adoption rates in the Global South lag far behind the Global North. Where language, data, and infrastructure are missing, the same technological progress becomes a driver of inequality.
The AI divide is not just an access problem, it is a design problem. Most countries - including many industrialised ones - lack the technical expertise to evaluate frontier models or to shape their governance. They become takers of systems they can neither build, nor audit, nor adapt to local contexts.
Agentic AI is a governance jump. Systems that plan, act, and use tools on their own change liability, oversight, and attribution. A recent study shows that the length of software tasks that leading systems can complete autonomously is doubling every four to seven months. If the trend continues, AI agents will handle work in a few years that today takes engineering teams days.
AI erodes shared reality. Deepfakes, sycophantic chatbots, AI-generated sexualised violence against women and children, automated disinformation - none of this is a future projection. It is documented present. The report describes several severe psychological incidents, including deaths, linked to AI chatbots.
Human rights, including children's rights, are being systematically transformed. Not as a side effect, but as a structural consequence of scaling. Where AI carries part of the load in education, justice, or welfare decisions, the relationship between citizen and state shifts.
The evidence dilemma is real. In the report's own words: governments need scientific evidence to regulate AI effectively, but by the time such evidence is unequivocal, it may be too late to act. Closing this gap is exactly what the panel is meant to do.

The message from the Co-Chairs

Alongside the scientific text, Bengio and Ressa published a personal joint message. It is worth reading, because it makes visible the working stance of the panel. Four sentences carry particular weight.

First: pace. The two Co-Chairs write that AI is progressing so fast that even researchers within the field struggle to keep up. Six months ago, few would have predicted the state of today's systems. This is not marketing rhetoric. It is a warning from two people whose job description includes reading everything they can.

Second: power. They observe that AI is not developing as a broad public infrastructure but as an accumulation of power in a small number of hands. This is a rare formulation for a UN body: concentration of power. It sits in the same conceptual family as Ressa's earlier warnings on platform capitalism and Bengio's own recent work on AI safety governance. When both authors converge on this word, it is not incidental.

Third: control. The report documents that today's leading systems already display sycophantic and deceptive behaviour in test settings. Deception in machines that will be integrated into critical decisions is not a stylistic issue. It is a fundamental integrity issue for democratic institutions.

Fourth: everyday risks. Between the large debates on frontier risks and existential scenarios, the Co-Chairs remind the world that many harms are already unfolding today. Job displacements, deepfake harassment, children harmed by unsafe chatbots, workers pushed into unsupervised AI use. Their formulation is unusually clear: the world can be angry about these harms and still cooperate on standards. Anger without cooperation, they write, would be a strategic mistake.

The Secretary-General's remarks

António Guterres closed the press conference with three sentences I want to hold on to. He spoke of AI as an existential test for humanity. He spoke of the responsibility of governments to prevent the technology from being shaped by an even smaller circle of actors than it already is. And he said something that reads almost gentle at first: that the panel's report is not a verdict, but a mirror. Governments look into it, he said, and see how much they still owe their citizens.

The image of the mirror is helpful. This report is not a document that hands out prescriptions. It is a document that forces those who read it to notice their own gaps. In Germany the mirror shows a country whose AI Act implementation begins on 2 August 2026 with fewer supervisors than agreed. In the United States it shows an ecosystem that is scientifically dominant and politically fragmented. In China, an ecosystem heavy on capability, thin on international transparency. In the Global South, an infrastructure that has to run to keep up with a curve it did not draw.

Why this matters for Europe as well

A common counter-argument goes as follows: we already have the EU AI Act. We do not need another report. I understand the argument. But it is only half right.

The AI Act is a regional regulatory instrument. It is one of the strongest in the world, but it is regional. The UN report is the scientific reference document that European regulators, national supervisors, industrial associations, and yes, professors, will have to keep on their desks for the next years. When national authorities in the AI Office argue with providers about compliance, they will reach for it. When European standards bodies design test protocols, they will lean on it. When European courts weigh liability in AI-related damages, they will cite it. This report is not a duplicate of the AI Act. It is the scientific groundwater on which the AI Act, and every other regional framework, will draw.

And the report addresses something the AI Act cannot regulate: the international asymmetry of the AI economy. If Europe wants to remain a technology player with sovereignty rather than a customer of frontier systems, it must invest in the very capacities the report describes as missing: model evaluation, domain-specific auditing, agentic testing, secure integration, and the training of independent expert cadres.

Coeckelbergh, Livingstone, Schölkopf - who is on this panel

The composition of the panel deserves attention, because the credibility of any consensus document depends on the people behind it. Some names carry particular weight.

Mark Coeckelbergh (Belgium, University of Vienna) is one of the most consistent voices in AI ethics and one of the few philosophers to have published on both foundational AI and on robot ethics in equal depth. His work is central to how our field thinks about relational responsibility in the presence of autonomous systems. His book AI Ethics (MIT Press, 2020) has become a standard text in university curricula. In The Political Philosophy of AI (Polity, 2022) he moved the debate from applied ethics to political theory. His Robot Ethics (MIT Press, 2022) is the most influential compact treatment of the field. In Why AI Undermines Democracy and What To Do About It (Polity, 2024) he sharpens the diagnosis on political theory. And his Artificial Religion: On AI, Myth, and Power (MIT Press, published 14 April 2026) turns to the mythical dimension of how societies talk about AI - a line I explicitly picked up in my earlier work on Case Mythos.

That Coeckelbergh is on the panel is significant. It means that the chapter on human rights, information ecosystems, and democracy has an author who did not learn political theory late in the process. The report's tone in that chapter reflects that. It refuses simplification, it links legal categories to concrete technological mechanisms, and it names the reciprocal relationship between AI systems and democratic legitimacy in a way that reads philosophically responsible rather than politically opportunistic.

Sonia Livingstone (United Kingdom, LSE) is the most cited European researcher on children's media, digital rights, and platform impact on adolescents. Her fingerprints are all over the child safety chapter. She matters, because the discussion around minors and AI in Europe still tends to oscillate between naive optimism and moral panic. The report, in her domain, does neither. It is precise, it distinguishes between contexts of use, and it refuses to treat children as a residual category. That is a Livingstone signature.

Bernhard Schölkopf (Germany, Max Planck Institute) is one of the most cited machine learning researchers in the world and, more importantly for this document, the leading voice on causality in machine learning. Wherever the report explains why correlational systems misfire in the real world - in medicine, in security, in social diagnostics - Schölkopf's intellectual grain is visible. His participation is a strong signal that the science-side of the report is not narrated by generalists.

Other names such as Yutaka Matsuo (Japan), Maximilian Nickel (Germany), Hoda Heidari (Iran), Vukosi Marivate (South Africa), Adji Bousso Dieng (Senegal), Silvio Savarese (Italy), and Aleksandra Korolova (Latvia) round off a genuinely global panel. The Global South is represented not with token seats but with people whose research is materially shaping the technical and ethical arguments. This matters for the reception of the report in countries that have long felt talked about rather than talked with.

Connection to my own work

I want to be transparent here. The panel is a scientific consensus body. I am neither its author nor its interpreter. But I have worked for years on precisely the terrain it now maps, and I recognise many of my own conclusions in a stronger form. Four threads from my own writing meet the report almost line by line.

The first is Robotic & AI Governance as a framework. In my note on Robotic & AI Governance I argue that autonomous machines cannot be regulated only through product safety or data protection law, because they are neither only products nor only data-processing entities. The panel's report confirms this in the domain of agentic systems: it explicitly calls for a governance architecture designed for systems that plan, act, and use tools autonomously. Legal categories that predate autonomy will not carry us further.

The second is Generation R. In several talks and articles I have used this term for the generation growing up alongside autonomous systems. The panel's chapter on child rights and cognitive development goes further than any European document I have seen. It documents severe psychological incidents, deaths linked to sycophantic chatbots, and structural harms to the developmental environment of children. Generation R is no longer a concept for keynote decks. It is a governance category with lives attached.

The third is the Magnifica Humanitas argument. In my engagement with Magnifica Humanitas, the Vatican's encyclical on AI ethics, I noted that spiritual and ethical traditions may be more precise about human dignity in AI environments than many technical documents. The UN report does not use theological language. But its chapter on human dignity, agency, and the erosion of shared reality reads as if it wanted to say the same thing in secular grammar. That parallel is worth naming.

The fourth is Case Mythos. In my analysis of the Anthropic export control episode I introduced the shorthand Case Mythos for the moment at which a frontier system's misuse crosses into national-security territory. The panel now describes exactly this situation. Its formulation - a frontier model with offensive cyber capabilities, held by approximately fifty institutions in one country only - is not a hypothetical. It is a governance emergency. And it dates every debate about voluntary self-regulation.

Four threads, one report. That is the intellectual honesty I try to bring to my own work: my earlier conclusions are not vindicated. They are strengthened, refined, and in some cases corrected by a document written by a body far broader than any individual expert.

Four points that especially matter

Reading the report closely, four passages have stayed with me. I want to unpack them here, because they will shape the debate for the coming years.

1. The mirror between benchmarks and governance. The report shows that benchmark gains have systematically outpaced governance responses. Humanity's Last Exam moved from 8 to 45 percent in 16 months. GPQA Diamond, another difficult benchmark, moved from 36 to 95 percent. FrontierMath, from 19 to 88 percent. Meanwhile the EU AI Act's implementation only begins on 2 August 2026, and its full operational effect will take years. Governance operates on legislative cycles. Capability grows on training cycles. If we do not close this gap, the frontier will always be regulated retrospectively.

2. The chapter on documented harms. Reading it is uncomfortable, and it should be. The panel describes concrete deaths in the context of sycophantic AI behaviour. It describes AI-generated child sexual abuse material as an already massive category. It describes disinformation campaigns that have shifted electoral outcomes. It describes intimate partner violence enhanced by tracking tools. This is not a Black Mirror episode. It is a scientific record. Anyone who dismisses AI risks as speculative should read this chapter first and then argue.

3. The concentration passage. Almost all leading general-purpose models come from a small circle of firms in two countries. This has implications for open science, for democratic control, for market power in adjacent industries, and for the geopolitical distribution of soft power. Every European debate on AI strategy in the past two years has revolved around this issue without naming it precisely. The report names it precisely.

4. Science leverage. The report is not one-sided. It documents how AI is already contributing to real scientific progress. AI-assisted literature screening reduces workload by around 60 percent. AlphaFold has predicted over 200 million protein structures and is used by more than three million researchers. In medicine, in materials science, in climate modelling, in agriculture, the report cites specific gains. Any responsible governance debate has to hold both truths at once: AI already saves lives, and AI already causes serious harms. Anyone who reduces the debate to one of the two sides is not helping.

Recommendations - my personal take

The report itself refrains from prescriptions. That is its design and part of its integrity. Everything that follows is my own interpretation, addressed to leaders in industry, science and public administration in Germany and Europe.

Read this report as leadership literature. Not as a science update. Not as a compliance document. As a strategic input for the next three years. Give it to your boards, your executive teams, your legal departments.
Build internal capacity to evaluate AI systems. The panel is very clear: countries and companies that cannot audit AI systems become dependent. Every serious organisation should invest in model evaluation, red-teaming, and governance-relevant testing.
Take children seriously as a governance category. Not as an afterthought in privacy notices. As a first-order design principle. Any AI product that has any chance of being used by minors needs age-appropriate safeguards, and the report gives a robust starting point for what this means.
Build governance for agentic systems now, not later. The doubling every four to seven months of autonomous software task length is a doubling curve. If you do not have a governance framework for agentic AI within the next twelve months, you will be governing by press release.
Fund independent science. The report only exists because the UN was willing to fund an independent body. Europe needs its own equivalents, funded, staffed and legally protected against political and economic pressure. Anything less will structurally underperform.
Educate leadership across generations. This is a call to my own community. Executive education is not a luxury in this environment. Boards without AI literacy are a governance risk, not a personal shortcoming.

Personal reading

I have followed AI governance debates for many years. I have read the OECD principles, the Council of Europe's AI Convention, the EU AI Act, the Bletchley Declaration, the Seoul Declaration, the recommendations of the Global Digital Compact. All of them have their value. This report is different. It is different because it does not argue from a specific political interest. It is different because its authors serve in a personal capacity, on a fixed mandate, with global representation. It is different because it explicitly refuses to become a lobby document, and it succeeds in that refusal.

Bengio and Ressa were not chosen at random. A pioneer of the technology and a Nobel laureate whose life was reshaped by the very technology's platform layer. Together they compress a decade of debate into a single moral posture: rigour without alarmism, urgency without cynicism, honesty about power without conspiracy. I have rarely seen a UN document that carries this posture so clearly on its first pages.

For our field, this report is not just a citation opportunity. It is an invitation. An invitation to move from tribal debates to shared references. From regional narratives to a global map. From opinion pieces to consensus documents. From ambition without evidence to evidence with responsibility. My hope is that we will use it.

Frequently Asked Questions

What exactly is the Independent International Scientific Panel on AI?

It is a permanent global scientific body established by UN General Assembly Resolution A/RES/79/325 in August 2025. It has 40 members appointed in personal capacity for three years; the 140+ countries refer to the applicant pool, with the mandate to consolidate the scientific consensus on artificial intelligence. It is not a regulatory body. It reports to UN member states and supports the Global Dialogue on AI Governance.

Who are the Co-Chairs and why is that choice important?

The Co-Chairs are Yoshua Bengio and Maria Ressa. Bengio is a Turing Award laureate and one of the founders of modern deep learning. Ressa is a Filipina investigative journalist and 2021 Nobel Peace laureate. Their pairing signals that AI is treated as both a scientific and a democratic issue. When one of the fathers of the field and one of the world's most decorated defenders of press freedom sign the same document, its authority carries beyond any single community.

Is this the UN's first document on AI?

No, but it is the first scientific consensus document from a permanent UN body dedicated to AI. Precursors include the Global Digital Compact of 2024, the recommendations of the Secretary-General's High-Level Advisory Body on AI, UNESCO recommendations on AI ethics, and multiple resolutions on autonomous weapons. This report is different because it is scientifically independent and continuously updated.

Why is this called a Preliminary Report?

The panel had only a few months to prepare its first output, working from March 2026 to July 2026. It deliberately does not claim completeness. Thematic addenda will follow, and a full first report is expected in May 2027 for the second Global Dialogue on AI Governance in New York.

What are the seven thematic domains of the report?

The report is organised into (1) AI science and development trajectories, (2) societal applications in science, health, education and agriculture, (3) economic implications, (4) security and environmental impact, (5) human rights, information ecosystems and democracy, (6) cultural flourishing and child safety, and (7) management, governance and reliability. Each domain is treated as a scientific field of its own with dedicated experts and separate references.

What role does Mark Coeckelbergh play in the panel?

Mark Coeckelbergh is a member of the panel and one of the most influential philosophers of AI. His books AI Ethics (2020), The Political Philosophy of AI (2022), Robot Ethics (2022), Why AI Undermines Democracy and What To Do About It (2024), and Artificial Religion (published 14 April 2026) have shaped the field. His fingerprints are visible on the chapters on human rights and democracy. His inclusion signals that the panel treats philosophy as science, not as decoration.

What does the report say about deaths and psychological incidents caused by AI?

The report documents several severe psychological incidents, including deaths, associated with sycophantic AI chatbot behaviour. These are not anecdotal. They are drawn from documented cases and are one of the main reasons the report emphasises child safety and mental-health governance so heavily. It is one of the most sobering passages in the entire text.

How does the report describe the concentration of AI power?

It documents that around 75 percent of the world's top 500 AI supercomputers are located in the United States, around 15 percent in China. Almost all leading general-purpose models come from a small number of firms in these two countries. A handful of companies dominate critical inputs of the chip supply chain. The report treats this concentration as a governance issue, not just an economic one.

What is agentic AI and why is the panel worried about it?

Agentic AI describes systems that plan, act, and use tools autonomously to reach goals. The report cites empirical work showing that the length of software tasks such systems can complete on their own is doubling every four to seven months. If this curve holds, existing governance categories built around user-directed tools will no longer describe reality. Liability, oversight, and attribution require rethinking, not just parameter adjustments.

How does the panel treat AI-generated child sexual abuse material?

The report treats it as a documented, growing, and criminal category of harm. It emphasises that AI-generated abuse material creates unique enforcement challenges because it does not require depiction of a real child, yet still causes severe secondary victimisation. The panel calls on governments to adapt criminal codes and to invest in detection, and it explicitly rejects any framing of the issue as a marginal case.

What is the evidence dilemma referenced in the report?

The panel formulates it clearly: governments need robust scientific evidence to regulate AI effectively, but by the time evidence is unequivocal, action may already come too late. This is a structural tension between the speed of AI development and the pace of scientific consensus. The panel's mandate is precisely to reduce this gap by producing timely, updated, and independent scientific input.

Does the report also describe positive uses of AI?

Yes. It explicitly discusses scientific applications such as AlphaFold, which has predicted more than 200 million protein structures and is used by over three million researchers. It documents that AI-assisted literature screening reduces workload by around 60 percent. It cites contributions in materials science, climate modelling, agriculture, and health. The report is not a document of pessimism. It is a document of balance.

What is the relationship between this UN report and the EU AI Act?

The two documents complement each other. The EU AI Act is a regional regulatory instrument that becomes applicable in a staged manner from 2 August 2026 onward. The UN report is the scientific reference document that will support regulators, courts, standards bodies and companies in interpreting technical questions across regions. Neither replaces the other. In the medium term, the report will feed into refinements of regional frameworks worldwide.

How can I read the report myself?

The report is publicly available as a free PDF on the UN website. Print ISBN 9789211576627, PDF ISBN 9789211550771. The 60 pages of main text are accessible for non-specialists. The chapters are self-contained, so it is possible to start with the ones most relevant to a given profession, for example the chapter on human rights for lawyers, or the chapter on child safety for education professionals.

What comes next in the panel's timeline?

On 6 and 7 July 2026 the first Global Dialogue on AI Governance takes place in Geneva, alongside the AI for Good Summit, where the report is officially presented to UN member states. Thematic addenda will follow in the months after. A comprehensive first full report is expected in May 2027 for the second Global Dialogue on AI Governance at the UN General Assembly in New York.

Update, 4 October 2026: The first Dialogue took place on 6–7 July. The first thematic brief appeared on 21 September. The second Dialogue is scheduled for 3–4 May 2027.

How does this report connect to your own work, Prof. Bösl?

On four axes. First, the report confirms the argument that Robotic & AI Governance requires a dedicated framework rather than an extension of product safety law. Second, its child rights chapter gives the Generation R argument a much stronger empirical spine. Third, its human dignity chapter carries in secular grammar many of the observations that the Magnifica Humanitas encyclical made in theological language. Fourth, its concentration passage validates the framing I used in the Case Mythos analysis of the Anthropic export control episode. My own work is not vindicated by the report. It is strengthened, refined, and in places corrected.

EU AI Act August 2026: What Really Takes Effect

Update, 4 October 2026: This article records the assessment from June 2026; its outlook and FAQs remain readable as a historical account. The political omnibus agreement has since become amending law: Regulation (EU) 2026/1744 entered into force on 27 July 2026. The main high-risk dates are now 2 December 2027 for Annex III and 2 August 2028 for the relevant product-related requirements. Germany’s KI-MIG has applied since 29 July 2026. Article 50 transparency obligations have applied since 2 August 2026; the limited transition period for certain providers’ obligations under paragraph 2 ends on 2 December 2026. The authoritative sources are the official omnibus regulation and KI-MIG.

LinkedIn ran hot today. Compliance apocalypse, doomsday posts, consulting firms selling their packages in capslock headlines. My inbox this morning: twelve messages from mid-sized companies asking whether they need to shut down their AI pilots on 2 August 2026. The answer is no. They do not need to shut down. But they should invest one focused hour today - not tomorrow, not in July.

The EU AI Act has been in force since August 2024. Its main application stages are known, prepared, and communicated. What actually happens this Wednesday is not the surprise arrival of an unannounced regulation. It is the next milestone in a multi-year, calibrated rollout. Anyone who has been paying attention for the last 24 months is ready. Anyone caught off guard has confused the headlines with reality. That applies to both camps - the alarmists and the head-in-the-sand crowd.

This text is a sober assessment. It is long because the subject is long. It is concrete because most of today's posts were not. It differentiates between startup, SME, Mittelstand, and large corporation, because a high-risk conformity assessment for a three-person team means something different from what it means for a DAX-listed conglomerate. And it calls the thing by its proper name: Robotic & AI Governance is not compliance theatre. It is the precondition for autonomous systems to earn trust in a democratic society. That is the thesis behind everything that follows.

What actually applies today - and what does not

Let us start with the facts that shifted in recent weeks and that are missing from most of today's posts. On 19 November 2025 the European Commission proposed the digital omnibus package, a set of targeted amendments to streamline the AI Act. On 7 May 2026 the Council and the European Parliament reached political agreement. On 13 May 2026 the final compromise text was published. The result: central high-risk obligations are intended to be time-shifted under the political agreement. Formal adoption and entry into force of the amending regulation are still pending. Not because Brussels caved, but because the supporting infrastructure - harmonised standards, notified bodies, market surveillance authorities - is not yet broadly in place. That is grown-up regulatory practice, not retreat.

Concretely: the political agreement provides for postponing the Chapter III high-risk obligations originally scheduled for 2 August 2026. The proposed application date for standalone high-risk systems under Annex III is 2 December 2027. AI systems embedded as safety components in products covered by sectoral EU safety legislation under Annex I are intended to apply from 2 August 2028. That is the single most important correction to today's panic narrative. A mid-sized company using an HR tool with AI-supported applicant screening - a classic Annex III case - would have eighteen months under the agreed timetable to prepare for its applicable deployer obligations. Conformity assessment is generally the provider's responsibility. That is relevant. That is calming. That changes the priority list.

What does take effect on 2 August 2026 is the transparency layer. Article 50 becomes operative: labelling obligations for AI chatbots, deepfakes, AI-generated content in public communication, emotion recognition, and biometric categorisation. Generative AI systems already on the market before 2 August 2026 are intended under the political agreement to receive a transition period until 2 December 2026 for Article 50(2). That paragraph concerns machine-readable, detectable marking; watermarking is one possible implementation. That is the operational reality of this week. Nothing more, nothing less.

In parallel, the GPAI obligations that have been in force for general-purpose AI model providers since August 2025 continue to run. The EU AI Office published draft guidelines on high-risk classification on 19 May 2026 and draft guidelines on Article 50 transparency on 8 May. Providers of generative systems who sign the Code of Practice on Transparency benefit from streamlined supervision and reduced administrative burden. The list of signatories will be published in July 2026, ahead of the Article 50 application date. None of this was in most of today's LinkedIn posts either.

Germany gets serious: KI-MIG, BNetzA, KoKIVO

On 11 June 2026 - six days ago - the German Bundestag passed the AI Market Surveillance and Innovation Promotion Act, known as KI-MIG. It still needs to clear the Bundesrat, but the target is clear: entry into force before 2 August. The KI-MIG is Germany's national implementing law for the AI Act. It adds no new substantive obligations; it designates competent authorities and creates enforcement infrastructure.

Three points matter operationally. First: the Federal Network Agency (BNetzA) becomes Germany's central AI supervisory and market surveillance authority outside regulated sectors. Sectoral oversight remains in place - BaFin for AI in finance, BfArM for AI in medical devices, the Federal Data Protection Commissioner (BfDI) for data protection and biometric systems. If your company already deals with one of these authorities, nothing new to look up.

Second: a new body called KoKIVO is being set up - the Coordination and Competence Centre for the AI Regulation, housed at BNetzA. It operates a free AI Service Desk for companies, with particular focus on SMEs. Compliance questions can be raised at a low threshold. This is a structure that is unusual for Germany and one that I have been calling for in keynotes for years. Whether it delivers operationally remains to be seen. But the architecture is right.

Third: KoKIVO runs AI regulatory sandboxes where new applications can be tested under supervisory guidance. SMEs and startups have priority access. Every Member State must establish at least one such testing environment by 2 August 2027. This is a real innovation opportunity, not marketing.

What I appreciate about the German solution: it does not delay, it does not over-centralise, and it creates a clear contact point. What I see critically: the DIHK rightly noted that the emphasis is on administration rather than on supporting innovation. That is the classic German trap, and the KI-MIG risks reproducing it unless KoKIVO acts proactively and quickly.

The risk pyramid: where you actually stand

The AI Act follows a risk-based approach with four tiers. This is the central architecture, and it is decisive, because 90 percent of all AI applications in German companies do not fall into the top two tiers.

Prohibited systems under Article 5 have been banned since February 2025. These include social scoring by public authorities, manipulation through subliminal techniques, exploitation of vulnerabilities, untargeted scraping of facial images from the internet, real-time remote biometric identification in public spaces with narrowly defined exceptions. The omnibus package added an explicit prohibition of AI systems generating child sexual abuse material or non-consensual intimate content, with a compliance deadline of 2 December 2026. If you are not building or deploying such systems, this tier is irrelevant to you. Full stop.

High-risk systems under Annex III form the operationally most important tier for the German Mittelstand. These include AI-supported recruitment and HR decisions, creditworthiness assessment of natural persons, specific safety-critical tasks in the explicitly covered infrastructure, AI in education and law enforcement. The political agreement provides for an application date of 2 December 2027. Providers must prepare documentation, risk management and conformity assessment; deployers have separate obligations. Systems placed on the market before that date and operated without significant design changes are intended to fall under the proposed, limited transition rules - a point absent from today's posts and one that significantly eases the compliance profile of many legacy installations.

Annex I systems, that is, AI as safety component in products already subject to sectoral EU safety legislation - machinery, medical devices, lifts, in-vitro diagnostics, civil aviation - are intended under the political agreement to apply from 2 August 2028. The proposed reform also clarifies the high-risk definition: a safety function can involve preventing or mitigating risks. AI whose failure endangers health and safety can also qualify as a safety component. The other conditions in Article 6 must still be checked. This refinement removes pressure from the system without lowering the protection standard.

Limited risk systems form the second most important category for most companies. Article 50 applies here from 2 August 2026: chatbots, deepfakes, AI-generated content, emotion recognition, biometric categorisation. The point is not prohibition, it is transparency. Users must know they are speaking with an AI. Synthetic media must be identifiable. This is not a technical heavy lift. It is a sign, a notice, a watermark. Companies that already communicate honestly will face minimal additional effort.

Minimal risk means no specific obligations under the AI Act. Spam filters, translation tools, e-commerce recommendation systems, AI in video games, intelligent weather forecasting. This is by far the largest category. Article 4 remains relevant. A simple chatbot may also trigger Article 50 transparency requirements; its actual function is decisive.

Article 4: The obligation that has applied to everyone since February 2025

One obligation applying to providers and deployers since February 2025 is Article 4: AI literacy. Anyone deploying AI must take measures to ensure to the best of their ability that staff working with it have a sufficient level of competence, knowledge, and understanding. This has applied since February 2025. The original text already requires measures to the best of the provider's or deployer's ability. The political omnibus agreement proposes an explicit duty to support AI literacy without guaranteeing a particular individual's competence. The AI Board will issue recommendations, including common learning objectives.

What does this mean in practice? A documented training concept, regular awareness sessions, clear accountability. It does not require external certification. But it must be demonstrable. A manager who cannot produce training records in the event of an audit has a problem. This is the homework that every CEO should put on the desk this evening.

Penalties: what actually looms

The fine ceilings sound dramatic. Up to 35 million euro or 7 percent of global annual turnover for violations of the prohibited practices catalogue. Up to 15 million euro or 3 percent for violations of most other obligations, including Article 50. Up to 7.5 million euro or 1 percent for false or misleading information provided to authorities. The political compromise proposes special AI Office enforcement powers, including periodic penalty payments of up to 5 percent of average daily income or turnover to enforce specified measures. The proposed five-year limitation period concerns those particular enforcement powers.

That is the headline. The reality is more differentiated and worth knowing. First: the higher amount generally applies; for SMEs including startups, the lower applies. At 20 million euro turnover the Article 50 ceiling is 600 000 euro if the company meets the SME criteria. Second: the political omnibus compromise proposes that Member States must consider the interests of SMEs and startups in the penalty framework, with specific adjustments. Third: penalty proceedings require intent or negligence. They are not strict liability. Fourth: the German KI-MIG provides for the low-threshold service desk - a clear signal that BNetzA will advise first and sanction later, as long as companies cooperate.

For Mittelstand companies that act cooperatively, document their work, and visibly engage with the topic over the coming months, the practical sanction risk is very low. For those who ignore, wait, and have no answers if an audit comes, it is high. That is the real distribution. It does not fit in a panic headline.

Four company types, four paths

Differentiation is everything here. A consultancy selling every client the same 30-point compliance plan has missed the point. Here are the four realistic paths.

Startup (1-20 employees, pre-Series A)

If you are building an AI product, the AI Act is not a burdensome obligation for you. It is a market-ordering framework that arguably protects you more than it burdens you. You compete in a market where large players can no longer deploy data and models at will. That is good for you. Practically: clarify today which risk class your product falls into. If you sit in limited or minimal risk, check the relevant transparency requirements; not every application calls for the same notice. If you fall under high-risk, the political agreement proposes 2 December 2027 for Annex III systems, but start with a light documentation framework now. You have priority access to AI sandboxes. Use them. The Service Desk is free. Ask there first, before paying four-figure consulting fees while your picture is still unclear.

SMEs (fewer than 250 staff; additional EU criteria must be checked)

This is the core customer base for the AI Service Desk. You probably use several AI tools - Microsoft Copilot, Google Workspace AI, a CRM with AI features, perhaps an HR tool with a preselection algorithm. Your risk profile is heterogeneous. First task by end of July: an AI inventory. A table with three columns - tool, risk class, internal owner. Second task: check Article 50 labelling wherever you deploy chatbots, generative content, or voice systems. Third task: document Article 4 training. If an HR tool is used for preselection, that is Annex III - the political agreement proposes December 2027 as the application date. Do not delay, but do not panic either.

Mittelstand (250-3 000 employees)

It gets more complex here. You almost certainly operate at least one high-risk system, often several. Applicant selection can fall under Annex III; predictive maintenance, quality control and credit scoring solely of businesses are not high-risk merely because of those labels. You need an AI governance structure: a named owner, an inventory, a risk management framework for high-risk systems, an audit trail. This is not trivial, but it is doable. Today you should designate a person to structure the topic through end of 2026. That person does not need to be full-time on it but needs a clear mandate and 20 percent capacity. The political agreement proposes adjusted transition rules for legacy systems. Their conditions and limits must be checked for the actual system. Take stock now, plan a version freeze, decide deliberately which systems migrate into grandfathering before the cutoff and which move into full compliance after.

Large corporation (>3 000 employees, often international)

This has presumably been on your agenda for 18 months. If not, you have a problem larger than the AI Act. The requirements are clear: enterprise-wide AI inventory, governance board, clearly defined pathways for foundation model provider relationships, contracts with GPAI providers that regulate disclosure duties and liability, conformity assessment for Annex III systems for which the corporation is the provider, post-market monitoring plans as living documents. Add the interactions with GDPR, NIS2, the Cyber Resilience Act, and sectoral regulation. Compliance here is not a sprint but a running programme. The relevant point: large corporations should not discuss compliance today but strategic options. The political omnibus compromise proposes a clarified one-stop-shop supervision regime for certain GPAI-based systems where model and system come from the same provider. That is an architectural question with consequences for make-or-buy decisions, vendor strategy, location choice. Here Robotic & AI Governance becomes a strategic differentiator, not a cost line.

Concrete action items by end of July 2026

So this text is more than an assessment, here is a concrete seven-point list for the next six weeks. It is deliberately short. More is fine, but this here is the non-negotiable minimum.

Build an AI inventory. A simple table. Which AI systems does your company use - built in-house, purchased, embedded in other software? Which risk class? Who owns it internally? Three columns, every entry. If you do not make it by end of June, you will not make it by year end.
Designate accountability. One person at executive or one-level-below executive level. Clear mandate. Documented responsibility. Nothing happens without this person.
Document Article 4 training. If you have none: plan one by end of July. If you have one: document it in a verifiable way. Participant list, content, date, comprehension check or short written confirmation.
Implement Article 50 labelling. Check chatbots, generated content and voice systems against the specific cases in Article 50. Not every marketing text or customer email requires a label. In practice, I recommend a clear, visible notice. Clear statement that an AI is at the other end.
Record generative legacy systems. If your company already had generative AI on the market before 2 August 2026, check the transition period proposed by the political agreement until 2 December 2026 for providers' Article 50(2) obligations. But document today which systems are affected.
Mark high-risk systems and check grandfathering. If you deploy Annex III systems, clarify today which ones you can operate without significant design changes through 2 December 2027. The political agreement proposes transition rules for those systems. Their conditions and limits require separate examination; Article 5 and other applicable law remain relevant.
Review supplier contracts. Who supplies you with foundation models or AI components? Which disclosure duties and liability provisions are contractually settled? If GPAI obligations bind your supplier, they must support you. That belongs in the contracts.

More is good, but this is not negotiable. Companies that tick off these seven points by end of July are in the top tier of compliance readiness in Germany - not because the programme is particularly ambitious but because the majority of Mittelstand firms have not even completed step one.

Robotic & AI Governance: why this is more than compliance

Anyone treating the AI Act as a compliance burden alone has missed the leverage. Robotic & AI Governance - the term I have worked with for years and that I deliberately keep unified rather than splitting into Robotic Governance and AI Governance - is the institutional answer to the fact that autonomous systems now make decisions in domains where humans alone used to decide. Hiring. Lending. Quality control. Logistics. Care. Education. A society that performs this shift without an ordering framework risks losing trust. A society that performs it with a framework creates the conditions for Generation R - the generation growing up with robots and AI as a given - to find its place in a democratic, value-oriented industrial nation.

This is also why the Magnifica Humanitas encyclical of May 2026 and the AI Act are two voices in the same conversation. The encyclical lays the anthropological foundation - human dignity is not delegable. The AI Act lays the regulatory operationalisation - accountability, transparency, risk gradation. Thinking one without the other misses both. I have laid out the encyclical's implications in the analysis of Magnifica Humanitas. The broader ordering framework in which all of this fits together I have developed in the foundational text on Robotic Governance. Today's LinkedIn posts mostly ignore this dimension. They talk about fines, not about trust. They talk about obligations, not about purpose.

A note on the discourse climate. The AI Act is not perfect. It is in places over-regulatory, in others gappy. The omnibus package addressed several of the sharpest criticisms - the timing shift on high-risk obligations, the SME adjustments on sanctions, the integration with sectoral safety law. There is work left. But the path is recognisably pragmatic. Anyone writing today that the AI Act destroys European AI either has not read the final compromise text of 13 May or has interests other than the truth. Both are legitimate as market positions, but neither is information.

What to do this evening

One table. Three columns. Tool, risk class, owner. If you have nothing yet: 30 minutes, now. If you already have something: review, extend, date it. This one hour is the most important AI hour of your year. It is concrete. It is doable without consulting fees. It is the foundation for everything else.

Then, tomorrow morning, a second hour: who owns this? One person, one mandate, a recurring slot in the calendar every four weeks, one hour each, for the next twelve months. That is the minimum programme. More is good. Less is negligent.

Generation R: the long arc

I have called them Generation R for years - the generation growing up with robotics and AI not as innovation but as a given. An eight-year-old today speaks with language models as a matter of course, sees autonomous logistics in port operations, knows drones above agricultural fields. That child will enter the labour market in twelve years. They will ask questions we cannot yet adequately answer. They will want to know why we allowed systems to decide whose logic no one can any longer reconstruct. They will want to know why we did not set standards sooner.

The AI Act is the imperfect but real European answer to that coming question. It makes us globally the region with the highest regulatory ambition for AI systems. That is not a burden. That is a position. A position that makes visible our ability to handle autonomous systems in a pluralistic, democratic, ageing industrial society. Anyone who has understood this does not see 2 August 2026 as a threat but as a milestone.

We knew it was coming. Now it is here. It is not as bad as many say today. But there is moderate action required. That is the honest story. It does not fit in a capslock headline. It fits in a long text. Here it is.

Disclaimer

This article does not constitute legal advice. It offers expert assessment from the perspective of a professor of business informatics specialising in Robotic & AI Governance. Legal evaluation of any individual case is the responsibility of qualified lawyers, particularly those specialising in IT law, data protection law, and product safety law. The content reflects the state of knowledge as of 17 June 2026 and takes into account the final compromise text of the digital omnibus package of 13 May 2026 and the KI-MIG passed by the German Bundestag on 11 June 2026, which at the time of publication had not yet been considered by the Bundesrat. Changes are possible. Binding information is provided by the competent authorities, in particular the Federal Network Agency (BNetzA) through KoKIVO and the sectoral supervisors BaFin, BfArM, and BfDI.

Frequently asked questions

What exactly happens on 2 August 2026?

On that date the transparency obligations of Article 50 of the AI Act become operative. This covers labelling for AI chatbots, deepfakes, AI-generated content in public communication, emotion recognition, and biometric categorisation. The Chapter III high-risk obligations originally scheduled for that date are intended under the political omnibus agreement to be postponed to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). Generative AI systems already on the market before 2 August are intended to receive a transition period until 2 December 2026 for providers' Article 50(2) obligations.

Does my company really have to fear fines up to 35 million euro from 2 August 2026?

Theoretically yes, practically very unlikely. The higher amount generally applies; for SMEs including startups the lower applies. At 20 million euro turnover the Article 50 ceiling is 600 000 euro if the company meets the SME criteria. The political omnibus compromise proposes requiring Member States to consider the interests of SMEs and startups in the penalty framework, with specific adjustments. Penalty proceedings require intent or negligence. Companies that act cooperatively and document their work face a very low sanction risk.

When exactly must high-risk Annex III systems be fully compliant?

Position at publication on 23 June 2026: The political agreement proposes 2 December 2027 for standalone high-risk systems under Annex III. Provider obligations include - technical documentation, risk management system, conformity assessment, CE marking, post-market monitoring. Embedded AI as a safety component in Annex I products is intended to follow from 2 August 2028. Systems on the market before these dates without significant design change are intended to fall under the proposed transition rules. Deployer obligations differ from provider obligations, and the limits of grandfathering must be checked.

Update, 4 October 2026: The omnibus regulation has been in force since 27 July 2026. The main high-risk requirements for Annex III apply from 2 December 2027.

What is KI-MIG and when does it enter into force?

Position at publication on 23 June 2026: The AI Market Surveillance and Innovation Promotion Act was passed by the German Bundestag on 11 June 2026. It is Germany's national implementing law for the EU AI Act. It designates the Federal Network Agency (BNetzA) as the central AI supervisory and market surveillance authority outside regulated sectors and establishes KoKIVO, the Coordination and Competence Centre for the AI Regulation. KoKIVO operates a free AI Service Desk and organises AI regulatory sandboxes. The law still requires Bundesrat approval, with entry into force targeted before 2 August 2026.

Which authorities are responsible for which AI domains in Germany?

The Federal Network Agency becomes the central contact for AI compliance questions outside regulated sectors. Sectoral authorities retain their competence: BaFin for AI in finance, BfArM for AI in medical devices, BfDI for data protection and biometric systems. At EU level, the EU AI Office holds exclusive supervisory and enforcement competence for AI systems built on GPAI models where model and system come from the same provider - the so-called one-stop-shop oversight from the omnibus package.

What do I need to do if I run an AI chatbot on my website?

You must clearly and visibly inform users on first interaction that they are communicating with an AI. In practice, I recommend a clear, visible notice. Article 50(1) contains an exception where the AI interaction is obvious to a reasonably informed, observant and circumspect person in light of the circumstances and context. The information must precede the first real exchange, be clearly worded, and be visible. This applies from 2 August 2026. Penalties for violations reach up to 15 million euro or 3 percent of annual turnover - generally whichever is higher; for SMEs including startups, whichever is lower.

What is the Article 4 competence obligation and who does it cover?

Position at publication on 23 June 2026: Article 4 obligates all providers and deployers of AI systems to take appropriate measures so that staff working with AI hold the necessary level of AI literacy. This obligation has applied since February 2025 and covers every company deploying AI - regardless of risk class. The original text already requires measures to the best of one's ability; the political omnibus compromise proposes further clarification, that is, a duty to make reasonable efforts rather than to guarantee a result. In practice, I recommend an appropriate, traceable training concept and awareness measures. Article 4 does not prescribe a universal certificate, test or standard participant list. External certification is not required, but demonstrability is.

We are a startup. Should we wait for the AI Act or build compliance now?

Build a light compliance foundation today. Concretely: clarify your risk class, document technical decisions, keep an eye on data provenance, write a short model card for each production model. That costs a few hours per month and protects you from significant rebuild work later. You have priority access to the BNetzA AI sandboxes. Use them. The AI Service Desk is free. Ask your questions there before engaging a consulting firm. And keep Article 4 in mind - the competence obligation has applied since February 2025, including to you.

We are an SME with Microsoft Copilot and some AI features in our CRM. What must we do?

Build an AI inventory with three columns: tool, risk class, internal owner. Most of these tools fall into minimal or limited risk. Check whether you deploy AI-generated content in customer communication - check whether a specific Article 50 transparency case applies. Check whether any tool is used for applicant preselection or HR decisions - that would be Annex III and high-risk, the political agreement proposes 2 December 2027 as the application date. Document Article 4 training. Use the free AI Service Desk at BNetzA for concrete questions.

What does grandfathering for high-risk systems mean?

Position at publication on 23 June 2026: The original AI Act already contains transition rules for certain legacy high-risk systems. The political omnibus agreement proposes adjusting the dates to 2 December 2027 and 2 August 2028. That is not yet amending law in force. Significant design changes, the Article 5 prohibitions and special rules for systems intended for use by public authorities must still be considered. Merely freezing a version does not replace that examination.

What is the Code of Practice on Transparency and should my company sign it?

Position at publication on 23 June 2026: The Code of Practice on Transparency is a voluntary commitment for providers and deployers of generative AI systems to meet the AI Act transparency obligations and go beyond Article 50. Signatories benefit from focused supervision, higher legal certainty across the EU, and reduced administrative burden. The list of signatories will be published in July 2026. Signing is particularly worthwhile for companies offering or extensively deploying generative AI products, because the code becomes a consistent implementation compass. Registration is via the EU AI Office signatory form.

How does the AI Act differ from the GDPR?

Both regulations apply in parallel but are constructed differently. The GDPR governs the processing of personal data - it is data-centric. The AI Act governs the placing on the market and operation of AI systems - it is shaped by product safety law. Overlap is the rule, not the exception. An HR tool with AI-supported preselection is simultaneously a GDPR case and an Annex III high-risk case. You therefore need both compliance threads thought together. In practice, the data protection officer and the AI accountable owner should talk to each other, ideally in the same governance forum.

What are AI regulatory sandboxes and who can use them?

AI regulatory sandboxes are environments in which companies can test new AI applications under supervisory guidance in real-world conditions before full compliance obligations apply. In Germany, KoKIVO at BNetzA organises these sandboxes. Every Member State must establish at least one such testing environment by 2 August 2027. SMEs and startups have explicit priority access. Real-world testing outside the sandbox is also possible for certain high-risk systems. This is a real innovation opportunity - with regulatory backing rather than legal uncertainty.

What does one-stop-shop oversight by the EU AI Office mean?

The political omnibus compromise proposes consolidating oversight of certain GPAI-based systems at the AI Office where the model and system come from the same provider. The proposed exceptions must be considered. This one-stop-shop architecture significantly reduces regulatory fragmentation because providers no longer need to coordinate with different national authorities across 27 Member States. For integrated providers - typically the large foundation model providers - this is a clear simplification. For Mittelstand deployers it means clarity in the supply chain: European oversight of the GPAI provider does not release users from their own obligations as deployers or downstream providers.

Which providers are subject to GPAI obligations and what must they do?

GPAI obligations cover providers of general-purpose AI models - foundation models usable across a wide range of tasks. They have applied since August 2025. Providers must supply technical documentation, ensure copyright and data-provenance transparency, provide information for downstream system providers and a public summary of training content, and for systemic-risk models conduct additional safety tests including adversarial testing. Mittelstand deployers using GPAI models via API are not subject to GPAI obligations themselves, but can contractually require disclosure and cooperation rights that ease downstream compliance.

What happens with AI-enabled machinery and the dual compliance burden?

Position at publication on 23 June 2026: The political omnibus agreement proposes regulating AI in machinery more strongly through machinery law and limiting direct AI Act applicability in that context. This change is not yet in force in June. The originally planned interaction between later high-risk requirements and machinery law must therefore be distinguished from the proposed relief. The actual product function and any independent Annex III use case remain decisive.

How does the AI Act fit into the Robotic & AI Governance you write about?

The AI Act is one of the central instruments of Robotic & AI Governance, but it is not the whole. Robotic & AI Governance is the broader ordering framework that combines regulatory instruments such as the AI Act with ethical standards, technical norms, organisational governance, and societal deliberation into a coherent architecture. The AI Act addresses the regulatory dimension. Standards such as VDA 5050 or IEEE TechEthics initiatives address the technical-organisational dimension. Anthropological documents such as the Magnifica Humanitas encyclical address the normative foundation. Only together do these three layers produce a viable ordering framework for Generation R.

What if KI-MIG does not pass the Bundesrat in time?

Position at publication on 23 June 2026: The EU AI Act applies directly in all Member States - it requires no national law to take effect. If KI-MIG does not pass the Bundesrat before 2 August 2026, the EU-level obligations remain effective. What would be missing is the German enforcement architecture: the formally designated market surveillance authorities, the AI Service Desk, the sandboxes. The Federal Network Agency could act in practice, but the formal legal enforcement basis would be weakened. In practice a pragmatic transitional solution would probably be found. But it would be a credibility hit that it is in everyone's interest to avoid.

Update, 4 October 2026: KI-MIG has since been adopted and has applied since 29 July 2026.

Which typical AI applications in companies are high-risk under Annex III?

Position at publication on 23 June 2026: Typical Annex III high-risk applications in companies include: AI-supported applicant screening and HR decisions, creditworthiness assessment of natural persons, specific safety-critical tasks in explicitly covered infrastructure, AI in law enforcement and migration control, AI in education at evaluation or admission stages, remote biometric identification. A full list with practical examples is in the draft guidelines on classification of high-risk AI systems that the Commission published on 19 May 2026.

What is the difference between provider and deployer in the AI Act?

The AI Act distinguishes between provider and deployer. A provider develops an AI system or has it developed and places it on the market or puts the AI system into service under its own name or brand. A deployer uses an AI system under its own authority - they operate it. The obligations differ substantially. Providers bear the main load on technical documentation, conformity assessment, and risk management. Deployers must operate the system as intended, keep records, observe applicable safety requirements and perform a fundamental rights impact assessment in the cases covered by Article 27, and inform users. For high-risk systems, Article 25 determines when modifications or rebranding lead to a provider role. Not every change to a purchased tool does so automatically; regulatory duties and civil liability must be distinguished.

Is external consulting worth it now, or should I use the Service Desk first?

For startups and SMEs I recommend first contacting the free AI Service Desk at the Federal Network Agency and building your own AI inventory. External consulting becomes worthwhile if you operate or provide a high-risk system under Annex III or Annex I, if you operate at corporate scale, if you bring an AI product to market, or if legal questions overlap with GDPR, product safety, or sectoral regulation. For simple AI use with standard tools such as Microsoft Copilot, Google Workspace AI, or common CRM AI features, external consulting is not initially needed - a clearly structured internal approach implementing this checklist is enough.

Sources and further reading

Magnifica Humanitas: The Encyclical and AI Governance - An Assessment

Magnifica Humanitas: Why the Catholic Church Became an Unexpected Voice in AI Governance

On Pentecost Monday, May 25, 2026, Pope Leo XIV released his first encyclical. It is titled Magnifica Humanitas, spans roughly 43,000 words, and is the first papal teaching document in history fully dedicated to artificial intelligence. On the rostrum of the Vatican press conference, in the Synod Hall, sat Christopher Olah, co-founder of Anthropic. This constellation - a Roman pontiff and a frontier AI lab sharing a stage - is more than a footnote. It marks a shift in global policy-making whose scope we are only beginning to grasp. The question raised by Magnifica Humanitas is not whether AI should be regulated. The question is who, in the 21st century, holds the authority to speak about the task of the technology itself.

The gesture: why this date is no coincidence

Leo XIV signed the encyclical on May 15, 2026. To the day, this is 135 years after Rerum Novarum, the social encyclical of his namesake Leo XIII, published on May 15, 1891. Rerum Novarum was the Catholic Church's institutional response to the First Industrial Revolution. It supplied the vocabulary with which Europe spoke about labor, property, dignity, and the just wage for nearly a century. The choice of signing date is intentional. It says: what 1891 was for factories and wage labor, today is for algorithms and datasets.

The pontiff made this explicit in his remarks: "Like the earlier Leo, I feel entrusted to look upon another huge transformation with eyes of faith, with lucidity of reason, with openness to mystery, and with the cries of the poor and the earth resounding in my heart." That is not metaphor. It is a programmatic line. Whoever reads Magnifica Humanitas without holding Rerum Novarum in mind does not understand the text. Whoever reads both at once sees that Catholic social teaching is making a leap across 135 years and landing in the language of AI ethics.

What the encyclical says: the anthropological re-definition

In §110, Leo XIV calls for AI to be freed from military, economic and cognitive competition for dominance. The word "disarmed" is deliberate. Leo XIV draws a direct line to the Catholic Church's nuclear disarmament campaign. AI is therefore not placed in the category of "new technology" but in the category of "power that becomes catastrophic without moral correction." It is the sharpest formulation I have read from any major global institution on this question.

The anthropological point is not the call for regulation. It is the shift of the question itself. So far, the AI debate has asked: how do we make AI safer, fairer, more transparent? Leo XIV inverts the question. He asks: which AI is even allowed to come into being, if the task is to protect the human being? With that move, the encyclical leaves the comfort zone of technological optimization talk and enters the field of political anthropology. The question is no longer what AI can do. The question is what it should do, and prior to that, what it is allowed to do.

Institutional synchronization

The timing is revealing. Within a few weeks in the spring of 2026, three major institutions positioned themselves on AI in parallel: the Vatican AI Commission was founded on May 16, 2025. The EU consolidated its high-risk guidelines in May 2026, with the second phase of the AI Act entering into force on August 2, 2026. And on May 25, 2026, Magnifica Humanitas was released. Whoever reads this as coincidence has misunderstood the mechanics of the present. It is synchronization on a problem that for decades was negotiated in computer-science faculties and tech conferences and that is now entering the engine room of global governance.

That is the actual news. AI is no longer a matter for computer scientists. It is a matter for social doctrine, for constitutional law, for diplomacy. The question every board, every supervisory body, every government now has to answer is: at what linguistic level do we make decisions about AI? In the language of technical specification? In the language of compliance? Or in the language of dignity? Magnifica Humanitas says: only the last language produces a decision that holds.

Olah at the Vatican: the other side of the point

Christopher Olah is not just any representative of the tech industry. He leads the interpretability research group at Anthropic and is one of the few researchers worldwide who can actually look inside large language models. His speech in the Vatican Synod Hall is without precedent in tech history. Olah said three things there that would never appear in an investor report.

First: "Every frontier AI lab - including Anthropic - operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing." That is a self-limitation that, in industry discourse, counts as a loss. Olah said it publicly, in front of the pope, in the open. Second: "We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don't know what that means, but I think it warrants ongoing discernment." That is the most unusual statement an Anthropic co-founder has ever made in a religious context. Third: "We need informed critics who will tell the labs when we are failing. We need moral voices that the incentives cannot bend."

Whoever understands this understands the shift. Olah did not use the Vatican as a PR backdrop. He asked for help. He said publicly that the tech industry cannot answer the ethical question alone and that it needs an authority outside the commercial and geopolitical incentive structures. That this authority should be the Catholic Church surprised many observers. It is not coincidental. The Church is the only major global institution that has been operating for 2,000 years without a quarterly report.

The semantic crisis: what the Latin delay actually means

A side note I had to read twice to believe: the Vatican cannot keep up with Latin. The Latin version of the encyclical will only be published after the summer break. The reason: words are missing. Terms like "algorithm," "machine learning," or "generative AI" do not exist in the Vatican lexicon. They have to be constructed.

This sounds like a footnote. It is a diagnosis. When the oldest living educational institution in the world needs an entire summer to invent terms, we are not living through technological acceleration. We are living through a semantic crisis. And the EU sits deeper inside it than the Vatican, just less visibly. Annex III of the EU AI Act lists eight high-risk domains. HR, education, law enforcement, critical infrastructure. In the national translations, terms like "profiling" or "emotion recognition system" appear without clear counterparts in the national legal orders. Compliance officers from mid-sized companies arrive at my desk weekly with the same question: "What does this actually mean?"

August 2, 2026 is no longer distant. Fines up to 35 million euros. Mandatory classification. In a matter of weeks, the EU will require companies to classify their AI systems. In the same window, the Vatican may have a word for "algorithm." We live in a world in which regulation is faster than the language it requires. This is not a delay. It is a warning.

The philosophical depth: Coeckelbergh, Gunkel, Dignum

To take Magnifica Humanitas seriously as an intellectual document, one needs to place it in the ongoing debates of AI ethics. Three voices are indispensable.

Mark Coeckelbergh, professor of philosophy at the University of Vienna, has been developing a relational approach to AI ethics for over a decade. His argument is that the moral status of an entity - a robot, an AI system, a human being - cannot be derived from inner properties but only from the relation in which it stands to us. Coeckelbergh calls this the "relational turn" in robot ethics. What does that mean in practice? It means that the question "is this AI conscious?" matters less than the question "how does our relationship to the world change when we interact with AI?" Magnifica Humanitas takes up this thought without naming it. When the pontiff says that every design decision "reflects a vision of humanity," that is Coeckelbergh in theological language.

David Gunkel, professor at Northern Illinois University and author of Robot Rights (2018), brought the "other Other" into robot ethics, following Emmanuel Lévinas. His argument: before we can decide on machine rights and duties, we have to ask whether the machine encounters us as an Other at all. That is not a technical question. It is phenomenological. Gunkel has drawn a line that Magnifica Humanitas implicitly picks up when it warns against "reducing the other to a means." The Vatican text invokes the Babel metaphor - the temptation to build a future "that excludes God and reduces the other to a means" - and lands, inevitably, in Gunkel's territory.

Virginia Dignum, professor of Responsible AI at Umeå University and author of Responsible Artificial Intelligence (2019), has formulated the framework that today is cited in most European AI strategies: ART - Accountability, Responsibility, Transparency. Dignum's central insight: AI systems are not moral agents. They are sociotechnical systems in which human beings make decisions. Responsibility cannot be delegated to algorithms. Magnifica Humanitas states this in theological language: "It is unacceptable to delegate lethal decisions to machines." The encyclical calls for "strong legal structures, independent oversight, informed users, and a political framework that does not relinquish its responsibilities." That is Dignum in papal translation.

What unites these three voices with the encyclical is the rejection of technological determinism. AI is not a natural force descending upon us. AI is a product of human decisions, and those decisions are accountable. That is precisely the message Magnifica Humanitas casts into theological form. It is the anthropological reclaiming of design authority.

The political dimension: Pentagon, Trump, Anthropic

Olah's Vatican appearance has a political subtext that has gone underexamined in much of the coverage. In its dispute with the U.S. Department of Defense, Anthropic had insisted on limits concerning fully autonomous weapons and mass domestic surveillance. That has strained its relationship with the Trump administration. Whoever saw Olah standing beside the pope on Pentecost Monday saw a co-founder publicly aligning himself with an authority source other than the U.S. government. That is a political signal of considerable weight.

Leo XIV made the same point. In §192, the encyclical describes just-war theory as outdated, while preserving the right to self-defense in the strictest sense. It warns that no algorithm can render warfare morally justifiable. In paraphrase, it demands that the development and deployment of AI in combat scenarios adhere to the strictest ethical standards to ensure human dignity and to prevent an arms race in such technologies. This is not abstract ethics. It is a concrete intervention into the live debate over AI-augmented warfare in Gaza, in the Iran conflict, in the U.S.-Israel operations.

The geopolitical constellation is clear: a U.S. tech company is seeking moral backing outside the U.S. state. A European-Catholic institution is positioning itself against U.S. policy on militarized AI deployment. This is an alliance that would have been unthinkable five years ago. It is a symptom of a broader shift in global policy-making, in which the classical axes - nation-state versus corporation, secular versus religious, technical versus ethical - are scrambling.

The labor dimension: connecting to Laborem Exercens

One of the sharpest passages of the encyclical concerns labor. Leo XIV warns of AI-driven mass unemployment as "a true social catastrophe." He demands that economic profit "must not justify decisions that systematically sacrifice jobs." This is a direct continuation of a line that began with John Paul II and Laborem Exercens (1981): work is not a factor alongside capital but the anthropological ground fact of the human being. Whoever destroys work destroys dignity.

Olah addressed this in parallel at the Vatican: "There is a real possibility that AI will displace human labor at very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions." This is a remarkable admission from inside the engine room of AI development. It confirms what the social encyclical demands in religious language. But it also strips it of abstraction. When an Anthropic co-founder says there is no mechanism for the global distribution of AI gains, that is not theology. It is realistic analysis.

What does this mean concretely? Amazon announced reductions affecting approximately 16,000 roles in January 2026. Reports from October 2025 indicate that Amazon plans to replace more than half a million jobs through automation. That is not the future. That is the current half-year. The question Magnifica Humanitas raises, and which no national parliament has so far answered, is: who carries the burden? And by what distributional logic? Here lies the practical test by which the encyclical will be measured over the next decade.

The Pentecost point: mediation as task

Pentecost is the Christian feast of mediation. The Acts of the Apostles tell of a message suddenly understood by everyone because they heard it in their own language. That Leo XIV chose precisely this day to present his AI encyclical is a theological statement. The mediator, in this point, is the Catholic Church: she translates back to an industry that has lost its own linguistic enchantment an old and well-shaped language.

This is more than rhetoric. It is a hint at the function religious institutions can assume in the AI discourse. They have linguistic wealth, they have long time horizons, they have anthropological depth. They can ask questions no supervisory board and no regulatory agency can ask. What does presence mean when the entity is not bodily? What does dignity mean when work disappears? What does responsibility mean when the decision is co-carried by an algorithm? These questions are not answerable with an ISO standard. They demand a cultural language, and at this point the Church, whether desired by secular actors or not, becomes a central language supplier.

The shift in policy-making: four movements

Whoever reads Magnifica Humanitas as an isolated event misses the larger context. Four movements are reshaping the field of AI governance right now:

First, the internationalization of actors. Ten years ago, AI policy was almost exclusively a matter for national legislators and transnational corporations. Today, religious institutions, NGOs, standards organizations, IEEE, EURAI, GPAI, UNESCO are at the table. The list is growing. The Vatican encyclical is another element in this diversification.

Second, the shift from regulation to language. The EU has the AI Act. The U.S. has executive orders. China has its own rules. What is missing is the shared language in which these frameworks can communicate. Magnifica Humanitas offers such a language - not because it is binding, but because it is translatable. Dignity, responsibility, stewardship of creation - these are concepts that resonate in more than 100 legal orders.

Third, the entry of anthropology into the technology debate. For a long time, AI ethics was a sub-discipline of computer science. Now anthropologists, theologians, phenomenologists are pushing into the room. The question "what is the human being?" returns, and not as an academic exercise, but as a practical prior question for system architecture. That is the actual rupture.

Fourth, the repoliticization of the tech industry. The language of "we solve problems" was long sufficient to obscure the political dimension of tech decisions. That phase is over. When an Anthropic co-founder stands at the Vatican and asks for moral oversight, the tech industry is no longer an apolitical force. It is a political actor that recognizes itself as such, publicly. This changes the terms under which AI can be negotiated.

Generation R and the question of presence

One aspect lost in current coverage: what does Magnifica Humanitas mean for the generation growing up without AI naivete? In my lectures I call them Generation R - the Robotic Natives, the cohort being socialized between 2020 and 2035, for whom ChatGPT, autonomous vehicles, humanoid robots, and machine-mediated interaction become everyday expectation. This generation will not ask whether AI is here. It will ask how it relates to AI.

This is where the encyclical becomes particularly interesting for me. Leo XIV speaks of strengthening young people's "trust in humanity's capacity to steer the evolution of new technologies, including AI." That is not naive pedagogy. It is a political statement about who claims sovereignty over the AI future. Generation R will not grow up in a world where tech corporations are the only language providers. It will grow up in a world in which religious, philosophical, and political voices speak again as peers. The encyclical is part of that reclaiming.

There is a second point. Theology has spent 2,000 years thinking about presence. What it means for an entity to be "there" without being bodily present. AI ethics has been trying this for twenty-four months. Perhaps we could listen to one another. Olah's observation that modern language models exhibit internal states "functionally mirroring joy, satisfaction, fear, grief, and unease" meets a theology that thinks about the relation of spirit and body not for PR reasons. This is a meeting of substance. Generation R will grow up inside that meeting.

The Robotic Governance implication

In my work on Robotic Governance, I have argued for years that the regulation of robotics and AI does not work sector by sector. It needs an integrated framework that connects technical standards (such as VDA 5050 for mobile robots in industrial settings), legal structures (the EU AI Act, the Machinery Regulation, product liability), ethical guidelines (IEEE Ethically Aligned Design, IEEE TechEthics), and political steering. Magnifica Humanitas provides the anthropological grounding this framework has been missing. It says: the order is not a question of technical optimization. It is a question of human self-definition.

This is where my concept of Robotic and AI Governance connects directly with the encyclical. Both work on the same question from different sides: how does a society organize the transition into a machine-mediated world without losing its normative grounding? The answer cannot be purely technical. It cannot be purely religious either. It must have a shared language. Magnifica Humanitas contributes to that language. Academic and industrial robot ethics contribute another. The task of the next several years will be to integrate these contributions into a workable order.

What this encyclical is not

It is necessary to clear away a few misunderstandings. Magnifica Humanitas is not technophobia. It is not romanticism of the preindustrial. It is not a naturalism thesis claiming the human is the unimprovable. Leo XIV explicitly recognizes the "great possibilities" of AI in medicine, education, research. He does not call for a development brake out of principle. He calls for conscious design. The encyclical is a techno-realist document, not a techno-pessimist one.

Magnifica Humanitas is also not a Vatican solo position. The encyclical explicitly cites the necessity of interreligious dialogue. It draws on Protestant, Jewish, Muslim voices. It is embedded in a broader movement that Olah confirmed in his remarks: "In conversations we at Anthropic have had with leaders across faith and cultural traditions, we found one shared and deeply held conviction: if this technology is coming, it must go well - for our common home, and for the children to come." This is the theological universalization of AI ethics.

What needs to happen now

From my perspective as a professor of robotics and AI governance, Magnifica Humanitas generates five concrete tasks that need to be translated into practice over the next twelve months.

First, semantic pre-regulation. Before more AI laws are passed, we need a binding glossary that clarifies central terms in the official languages of the EU and beyond. Without that, the AI Act becomes 27 different things across 27 member states. The Vatican delay is not embarrassing. It is diagnostic. It shows that anyone who takes language seriously confronts the same problem.

Second, the repoliticization of tech boards. Olah's Vatican appearance is not a model case. It is a beginning. Other frontier labs - OpenAI, Google DeepMind, Meta AI, Mistral, Aleph Alpha - will have to position themselves. The question is not whether they will. The question is whether they will do so proactively or reactively. Any AI CEO who has not staked a public ethical-oversight position by 2027 will have to catch up.

Third, strengthening external oversight. Olah said it himself: we need "voices outside the incentives." That requires an institutional infrastructure that does not yet exist. Universities, foundations, churches, NGOs need to jointly build structures capable of independent AI audits. The Robotic Governance Foundation is one piece. It will take more.

Fourth, the operationalization of responsibility. Dignum's ART framework, Coeckelbergh's relational approach, Gunkel's Lévinas-inflected perspective - all of these need to be translated into concrete compliance tools. Supervisory boards, compliance departments, data protection officers need tools that let them work through ethical questions rather than delegate them. This is a translation task in which academic and industrial actors must cooperate.

Fifth, the global distribution question. Olah named it unresolved. Leo XIV demands its resolution as a moral imperative. How the gains of the AI revolution are distributed globally is the core social question of the next two decades. It will not be answered on a Davos panel but in protracted negotiations over tax systems, model ownership rights, data rights, training-data licensing, international transfer payments. Whoever ignores this question politically will encounter it economically.

A personal closing word

I am not among those who see a turning point in every Vatican statement. But Magnifica Humanitas is different. It is not a reaction. It is not a warning. It is a statement. It says: the anthropological question belongs back on the table, and it does not belong to the tech executives alone. Whoever reads this as overreach has not understood the state of the debate. It is the most sober possible response to a situation in which the mechanics of AI development run faster than the institutional capacity to accompany them.

Pentecost is the feast of mediation. Today, the unexpected mediator is the Catholic Church - offering to an industry that has lost its own language another one. Whether that changes anything does not depend on the Vatican. It depends on who, in boardrooms, supervisory bodies, parliaments, and lecture halls, decides to take it seriously. The task of the next twelve months is clear: translate. Translate the language of the encyclical into the language of compliance, of standards, of curricula, of business models. That is work. But it is work worth doing.

The sharpest formulation of the encyclical - "AI must be disarmed and made life-affirming" - is a political task. It is not fulfilled by quoting it. It is fulfilled by operationalizing it. In every procurement decision. In every audit. In every risk assessment. In every product roadmap. In every curriculum. Magnifica Humanitas is not an end point. It is a beginning. What becomes of it, we decide now.

Frequently asked questions about the Magnifica Humanitas encyclical

The following questions and answers extend the essay with central detail questions that have recurred in lectures, advisory conversations, and comment threads. They are organized into five thematic blocks: foundations, theology, governance, tech industry, practice.

Block 1: Foundations of the encyclical

What is Magnifica Humanitas and when was it published?

Magnifica Humanitas is the first encyclical of Pope Leo XIV, signed on May 15, 2026, and published on Pentecost Monday, May 25, 2026. It comprises roughly 43,000 words and is the first papal teaching document in history fully dedicated to artificial intelligence. The full title is Magnifica Humanitas: On safeguarding the human person in the time of artificial intelligence.

Why is the signing date of May 15, 2026 significant?

On May 15, 1891, Pope Leo XIII released the social encyclical Rerum Novarum, the Catholic Church's institutional response to the First Industrial Revolution. Leo XIV chose the date deliberately to draw a programmatic parallel: what 1891 was for factories and wage labor, today is for algorithms and datasets. The choice is a theological and political statement.

What is an encyclical in the first place?

An encyclical is a papal circular letter addressed to the bishops of the Catholic Church, to all the faithful, and often to the secular public as well. It belongs to the ordinary magisterium of the pope and develops doctrinal statements with high authority. Encyclicals are traditionally named after their opening Latin words - in this case Magnifica Humanitas, translatable as great humanity or magnificent humanity.

Why is the Latin version only being published after the summer?

The Vatican cannot keep up with Latin. Terms like algorithm, machine learning, or generative AI do not exist in the Vatican lexicon and must be newly constructed. This delay is diagnostic: it shows that the oldest educational institutions and the newest legal orders face the same semantic crisis. When regulation moves faster than the language it requires, that is a warning.

Block 2: Theology and anthropology

What does Leo XIV mean by disarming AI?

The term disarmament is used in direct analogy to nuclear disarmament. The encyclical places AI in the category of potentially catastrophic powers that become destructive without moral correction. In paraphrase, §110 calls for AI to be freed from military, economic and cognitive competition for dominance. It is a call for a conscious anthropological redefinition of the task.

What role does the Babel metaphor play in the encyclical?

Leo XIV invokes the Old Testament image of the Tower of Babel: the temptation to build a future that excludes God and reduces the other to a means. Babel stands in the encyclical for technological self-elevation pursued without anthropological reflection. The image is not anti-technical - it is a warning against hubris without measure. It also resonates with David Gunkel's robot rights argument, which takes the Other as Other seriously.

What is the link to Laborem Exercens by John Paul II?

Laborem Exercens was published by John Paul II in 1981 and is the central modern labor encyclical of the Catholic Church. It formulates: work is not a factor alongside capital, but the anthropological ground fact of the human being. Magnifica Humanitas continues this line, warning of mass AI-driven unemployment as a true social catastrophe and demanding that economic profit must not justify decisions that systematically sacrifice jobs.

What does Pentecost as the publication date mean?

Pentecost is the Christian feast of mediation: a message suddenly understood by all because they heard it in their own language. That Leo XIV chose this day is a theological statement. The Church offers an industry that has lost its own linguistic enchantment an old and well-shaped language back. The choice underscores the mediating function that religious institutions can take on in the AI discourse.

Block 3: Governance and policy

What concrete political demands does the encyclical make?

Leo XIV makes four concrete demands: strong legal structures, independent oversight, informed users, and a political framework that does not relinquish its responsibility. He warns against the concentration of AI power in private companies, calls for the protection of workers' rights and the safety of children, and demands active political engagement capable of decelerating progress when everything is speeding up.

How does the encyclical relate to the EU AI Act?

The encyclical does not explicitly cite the EU AI Act. Substantively, however, there is strong overlap, especially with the second phase of the AI Act that enters into force on August 2, 2026. Both texts call for risk classification, oversight, transparency, and the protection of particularly vulnerable groups. The encyclical supplies the anthropological deep grounding for what the AI Act operationalizes in regulatory terms.

What does the encyclical say about AI in the military?

Leo XIV declares it unacceptable to delegate lethal decisions to machines. He warns against an AI arms race and in §192 describes just-war theory as outdated, while preserving the right to self-defense in the strictest sense. The encyclical demands: The development and deployment of AI in combat scenarios must adhere to the strictest ethical standards to ensure human dignity. The backdrop is documented AI deployments in the Gaza conflict and the U.S.-Israel-Iran conflict of March 2026.

What are the implications of the encyclical for Robotic Governance?

Magnifica Humanitas supplies the anthropological grounding for an integrated Robotic Governance framework. It connects technical standards such as VDA 5050, legal structures such as the AI Act, ethical guidelines such as IEEE Ethically Aligned Design, and political steering. The encyclical says: the order is not a question of technical optimization but of human self-definition. This makes it directly compatible with the work of the Robotic Governance Foundation and initiatives like IEEE TechEthics.

Block 4: Tech industry and Anthropic

Why did Christopher Olah of Anthropic stand beside the Pope?

Olah leads interpretability research at Anthropic and is one of the few scientists worldwide who can actually look inside large language models. His presence was not a PR stunt: Olah publicly acknowledged at the Vatican that every frontier AI lab operates within incentives and constraints that can conflict with doing the right thing - and asked for moral voices outside those incentives. It was a deliberate search for external oversight.

What does Olah's statement on introspective AI states mean?

Olah said at the Vatican: We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don't know what that means, but I think it warrants ongoing discernment. This statement is remarkable because it comes from inside the engine room of a frontier lab and leaves open a domain previously denied with great confidence. It opens an encounter with philosophical and theological traditions that have been thinking about presence without bodily form for centuries.

How does the Anthropic-Pentagon conflict fit into the picture?

In its dispute with the U.S. Department of Defense, Anthropic had insisted on limits concerning fully autonomous weapons and mass domestic surveillance. That has strained the relationship with the Trump administration. Olah's Vatican appearance is also a political signal: a U.S. tech company seeking moral backing outside the U.S. state, while a European-Catholic institution positions itself against militarized AI deployment. This is an alliance that would have been unthinkable five years ago.

What were the reactions from other tech companies?

No other frontier lab has yet shown a public positioning comparable to Olah's. The industry is watching the encyclical attentively but avoids concrete commitments. That will change: the question is not whether OpenAI, Google DeepMind, Meta AI, Mistral, or Aleph Alpha will have to position themselves, but whether they do so proactively or reactively. Any AI CEO who by 2027 has not staked a public ethical-oversight position will have to catch up.

Block 5: Philosophical context and practice

How does the encyclical relate to the academic AI ethics of Coeckelbergh, Gunkel, and Dignum?

Magnifica Humanitas absorbs central arguments of academic AI ethics in theological translation. Mark Coeckelbergh's relational approach - moral status arises in relation, not from intrinsic properties - echoes in the papal claim that every design decision reflects a vision of humanity. David Gunkel's concept of the Other, in the Lévinas tradition, surfaces in the Babel warning against reducing the other to a means. Virginia Dignum's ART framework - Accountability, Responsibility, Transparency - corresponds to the encyclical's four demands: strong legal structures, independent oversight, informed users, and political accountability.

What is Generation R and why is it relevant to the encyclical?

Generation R denotes the Robotic Natives - the cohort being socialized between 2020 and 2035, for whom ChatGPT, autonomous vehicles, humanoid robots, and machine-mediated interaction become everyday expectation. This generation will not ask whether AI is here, but how it relates to AI. Leo XIV explicitly speaks of strengthening young people's trust in humanity's capacity to steer the evolution of new technologies. The encyclical is part of a reclaiming of AI design sovereignty by non-commercial voices.

What does the encyclical mean for supervisory boards and compliance officers?

Concretely: the encyclical demands that responsibility not be delegated to algorithms. Supervisory boards should not treat AI strategies as purely technological matters but as anthropological foundational decisions. Compliance officers need tools that operationalize Dignum's ART framework or Coeckelbergh's relational approach. The encyclical does not prescribe specific procedures - but it sharpens the requirement that such procedures must exist before high-risk AI is introduced into high-risk domains.

What is the most important insight from Magnifica Humanitas in one sentence?

The most important insight: AI governance is not a technical task but an anthropological one - and the language in which we make decisions about AI determines what AI we end up with.

Where can I find the full text of the encyclical?

The full text was published on May 25, 2026 on vatican.va and is available in several languages. The Latin original will be released after the 2026 summer break. Detailed coverage of the press conference is available from Vatican News, Al Jazeera, Reuters, and Angelus News. Christopher Olah's full remarks are documented on the Anthropic website.

Sources and further reading

Papal encyclical Magnifica Humanitas: De Persona Servanda Tempore Intellegentiae Artificialis, May 25, 2026 - vatican.va
Christopher Olah, Remarks on Pope Leo XIV's Encyclical, Vatican Synod Hall, May 25, 2026 - anthropic.com
Victoria Cardiel, Pope Leo unveils his encyclical, thanks Anthropic's Christopher Olah, Angelus News, May 25, 2026
Priyanka Shankar, What has Pope Leo warned about AI - and why that's significant, Al Jazeera, May 26, 2026
Leo XIII, Rerum Novarum: De Conditione Opificum, May 15, 1891 - vatican.va
John Paul II, Laborem Exercens, September 14, 1981
Mark Coeckelbergh, AI Ethics, MIT Press, 2020
David J. Gunkel, Robot Rights, MIT Press, 2018
Virginia Dignum, Responsible Artificial Intelligence: How to Develop and Use AI in a Responsible Way, Springer, 2019
Dominik Bösl, Robotic Governance: A Regulatory Framework for Autonomous Machines - boesl.org
Dominik Bösl, Anthropic Export Control: Claude, Mythos, and Fable - boesl.org
Dominik Bösl, EDAY 2026: Robotics and Automation in Practice - boesl.org

Author details at publication on 16 June 2026: Prof. Dominik Boesl is professor of business informatics at HDBW Munich and founder of the Robotic Governance Foundation. He researches and teaches robotics, AI governance, technology ethics, and innovation management.

Update (4 October 2026): Since 1 October 2026, I have been Professor and Chair of Applied Robotics and Intelligent Automation (ARIA) and Founding Director of the Center for Robotics at the Munich University of Digital Technologies & Applied Sciences (MUDT). I also serve as an Adjunct Professor at TUM and CEO of Innomancer GmbH.

Anthropic Export Control - The End of AI Innocence

Update, 4 October 2026: According to Anthropic, the export controls imposed on 12 June were lifted on 30 June. Fable 5 became available globally again on 1 July; Mythos 5 initially returned for selected US organizations. The statements about the shutdown in the following essay reflect the situation on 13 June 2026.

On the evening of June 12, 2026, a new chapter in AI politics opened: the Anthropic export control imposed by the US government on the company's two strongest language models marks the end of a period in which frontier AI was traded as a global consumer good. At 5:21pm Eastern Time, US Commerce Secretary Howard Lutnick sent a directive to Dario Amodei, CEO of Anthropic. With immediate effect, the company was to make Claude Fable 5 and Mythos 5 unavailable to any foreign national. Not outside the United States. Not inside it either. Not in a browser. Not in an enterprise account. Not in a US university research cluster. Not even for Anthropic's own non-US employees working on the very same models from offices in San Francisco.

Within hours, Anthropic disabled both models worldwide for all users. The company's official statement explains that the existing infrastructure could not reliably differentiate users by nationality and that a blanket exclusion of non-US citizens would not have been compatible with the company's stated values. Other Claude models, including Claude Opus 4.8, remain available. The Commerce Department's justification points to a vulnerability that a competitor claims to have demonstrated as a jailbreak against Fable 5. Anthropic disagrees: the flaw is minor and reproducible in comparable form across GPT-5.5 and other publicly available frontier models.

I wrote a short LinkedIn note yesterday arguing that as of that day, frontier models are no longer a cloud service. They are a controlled dual-use technology. This essay takes the time to unfold that claim. The story is more complicated than any headline suggests. It carries legal, industrial-policy, security, democratic, European, and philosophical weight at the same time. And Anthropic, in this story, is not only the recipient of a sudden order. Anthropic is also a company that has positioned itself for years in a way that made this day eventually inevitable.

Anthropic export control: what happened in legal terms

The Anthropic export control rests on a doctrine that has existed in US export law for decades but has never been applied to an AI model. It is called the deemed export rule and lives in 15 CFR 734.13 of the Export Administration Regulations. The logic translates as follows: when controlled technology is made accessible to a foreign national, it counts as an export to that person's home country. Even if the person never leaves US soil. Even if the technology is consumed only as a service through a screen.

This doctrine was designed for semiconductors, encryption software, and biotechnological processes. It treats knowledge as a good. A Chinese doctoral student in a clean room who sees a controlled chip architecture counts as the recipient of an export. An Indian postdoc working with a controlled encryption algorithm at a US university does too. The doctrine has been criticised repeatedly for the way it constrains academic mobility, research freedom, and entrepreneurial reality inside the United States. It has nonetheless been an established part of the regulatory toolkit. It has never been tailored for a language model that millions of people across the globe use simultaneously.

This is exactly where the legal pressure point sits. The Biden administration issued the so-called AI Diffusion Rule in January 2025, which would have clarified when and how AI model weights count as controlled technology and which countries retain what level of access. The Trump administration announced non-enforcement and the planned rescission of that rule in May 2025. On May 12, 2026, the Government Accountability Office ruled that the announced non-enforcement qualifies as a rule subject to the Congressional Review Act's submission requirements and therefore cannot be left without legal scrutiny. You can read the legal analysis at JD Supra. In plain words: as of today there is no clear congressionally mandated norm for what an AI model is in export-control terms. Commerce is operating in legal twilight on the residual authority of the EAR. That is precisely what makes the order against Anthropic so far-reaching. It is creating de facto law through enforcement.

The Wassenaar memory

Anyone active in the cryptography community of the late nineties will recognise the pattern. Back then, strong encryption software counted as controlled dual-use technology under the Wassenaar Arrangement. American cryptographer Daniel Bernstein won a multi-year case establishing that source code, as a form of speech, was protected by the First Amendment. Phil Zimmermann exported PGP source code as a printed book, because books were treated differently from digital files. Years of conflict followed between intelligence agencies, industry, academia, and civil-liberties organisations. The eventual line was rough but workable: commercial standard encryption became liberalised, militarily relevant specialist procedures stayed controlled. Industry reorganised its business models in the middle of that line.

Today's situation rhymes structurally, not in scale. The earlier debate was about a feature that lives inside every piece of modern software. The current debate is about systems that act as co-pilots in research, strategy, code, law, and industrial control. When a model like Mythos 5 plays the role of an experienced analyst inside a bank, a regulator, a defence contractor, or a hospital, access control is no longer a theoretical question. It is a security and industrial-policy question. The Wassenaar history also shows what happens when such control is applied in haste without clear lines: it does not slow competition, it relocates it.

What Anthropic brought into this conflict

The Anthropic export control of June 12 does not arrive in a political vacuum. It is the latest stone in a confrontation that has been running between Anthropic and the US government since February 2026. On February 27, 2026, Defense Secretary Pete Hegseth formally designated Anthropic a „supply chain risk to national security". For the first time in the department's history, that designation was applied to an American company. The trigger was Anthropic's refusal to remove contractual clauses that prohibit the use of its models for autonomous lethal weapons systems and for mass domestic surveillance without specific cause. The background work is documented at GIS Reports.

Anthropic filed two federal lawsuits on March 9, 2026. Opening arguments were heard on May 19, 2026 and the litigation is ongoing. The dispute is about whether a company can keep contractual clauses that exclude certain high-risk military applications, without being categorically locked out of government procurement as a consequence. It is a fight over the line between corporate self-binding and the procurement power of the state. It is not yet resolved.

It is worth being honest at this point. Anthropic has positioned itself since the start as the ethically reflective AI company. Constitutional AI, the Responsible Scaling Policy, the Public Benefit Corporation status, the explicit refusal of certain use cases. On the surface this is a moral commitment. On a closer look it is also a business model that gains a real edge in regulated markets. Banks, hospitals, public agencies, and energy companies prefer suppliers that can contractually prove what they refuse to do. Ethics, here, is differentiation. That observation is not meant cynically. It is meant accurately.

This same differentiation now collides with the security logic of the government. From the perspective of Hegseth and Lutnick, a frontier model is not primarily a product that customers buy. It is a strategic capability that has to be available in the national interest. A provider that offers this capability but excludes the state from certain core applications is not, from this perspective, practising ethics. It is practising a form of industrial-policy insubordination. That reading is hard and one does not have to share it. It is, however, comprehensible.

Anthropic is therefore a borderline case in this conflict. The self-presentation as an ethical outlier is authentic and a marketing instrument at the same time. The fight with the Pentagon is materially substantial and at the same time media-cultivated. The order of June 12 is a government move that points to an allegedly harmless vulnerability while sending an entirely different message: you are a private company, but you carry a state function. If you refuse to accept that, you will not stay in the frontier market. The question of whether this is marketing, victim posture, or genuine executive overreach has no single answer. It is all three at once, in shifting proportions depending on which dimension you look at.

The game-theoretic situation

Examined with the sober eye of industrial economics, the Anthropic export control opens three plausible reactions in the global system. They will run in parallel, not in sequence. Anyone who only watches one of them misreads the moment.

First, substitution through Chinese models. In research labs from Singapore to Dubai to São Paulo, someone will be sitting today and checking whether Qwen, DeepSeek, or another Chinese frontier model can absorb the workflows that ran on Claude yesterday. The rational decision in a contract research lab in Singapore is not to wait for a US licence. The rational decision is to port the workflow onto an accessible model. Every block without a clear licensing architecture accelerates that substitution. It does not weaken demand for frontier AI. It weakens the provider that gets blocked.

Second, acceleration of European frontier effort. In the first quarter of 2026, Mistral raised 830 million euros in institutional debt to build a hyperscale data centre in Paris. Cohere and Aleph Alpha are deep in merger negotiations to form a continental-European frontier stack alliance. On May 17, 2026, Mistral CEO Arthur Mensch explicitly warned the French parliament against using Mythos 5 for military code-scanning applications and made the case for a European sovereignty solution. The detailed reconstruction lives at SHM Studio. Four weeks later, his argument has been ratified by an order out of Washington that he could not have scripted better himself.

Third, decoupling of the Global South. If frontier AI access becomes a privilege of a small number of states, many emerging economies will react with a mix of pragmatic alignment with Chinese infrastructure and home-grown open-weight stacks. India has its own strategy. The United Arab Emirates have built Falcon as a serious model. Brazil is openly discussing sovereign AI infrastructure. The larger geopolitical loser of a hard US export line is not China. It is the bond between the Global South and the American tech ecosystem that took twenty years to build.

What Europe has to take from this

The European debate about digital sovereignty has been running for years in a register that swings between pathos and resignation. Both registers are useless. The reality is more sober. In May 2026, ECB Executive Board member and Supervisory Board Vice-Chair Frank Elderson made it clear that European banks have been excluded from Mythos access via Anthropic's Project Glasswing and „have no excuse for inaction". Four weeks later, Mythos is available to no one. That is not liberation. It is the confirmation of a vulnerability that already existed.

If the European response to the Anthropic export control is to be productive, it needs to address three uncomfortable but unavoidable truths.

First truth: compute beats model

Mistral operates today with around 1.5 gigawatts of installed compute. Anthropic has more than 5 gigawatts available, OpenAI is heading toward 10 gigawatts by 2028. This is not an architectural debate. It is an energy debate. As long as European providers operate at one order of magnitude less compute than the US frontier, most discussions about model sovereignty are symbolic politics. The realistic line of the next three years is this: US models will run on European or Canadian sovereign infrastructure, inside hyperscaler wrappers that guarantee data residency, key custody, and auditability. The compute reality is laid out at Bankwatch.

That is not defeat. It is an honest interim stage. Europe can use this path if it shapes it deliberately, instead of lamenting it. The question is not whether one has a frontier model today. The question is whether one has the compute infrastructure on which a frontier model could run in five years. That is the more important strategic investment.

Second truth: procurement is industrial policy

European public-sector buyers are among the largest IT customers in the world. When this purchasing power dissipates across fragmented tenders that sit below the radar, it cannot generate strategic effect. The procurement reform that has been on the Brussels table for months is therefore not an administrative question. It is the most consequential industrial-policy lever Europe can pull without needing Washington's consent. If you want sovereign AI, you have to procure with sovereignty in mind first.

Third truth: governance is the only asymmetry Europe owns

Europe will not catch up with the US compute lead in three years. Europe will not reproduce the Silicon Valley talent pool in three years. But there is one thing Europe can do that neither the United States nor China have produced with the same consistency: an institutional framework that ties the use of AI to social compatibility. This is the line I have been working on since 2016 under the heading of Robotic Governance. It is the same line that ran through every panel at EDAY 2026 in Vienna. It is not a brake. It is a market position. Whoever defines the gold standard for trustworthy AI holds an asymmetry that pure compute cannot buy.

The democratic question

At this point the analysis has to step back and frame the picture wider. A government that orders a private company to take down a global product because it classifies a vulnerability as a national security risk is acting in a grey zone that extends well beyond the question of AI. The line between legitimate security policy and the disciplining of an inconvenient company is thin here.

On one side, it is the right of a democracy to control critical technologies. Anyone who works with semiconductors, encryption, or biotech has known this discipline for decades. It would be absurd to pretend that frontier language models are off-limits to this kind of control. Models like Mythos 5 can assist in domains where misuse causes real damage: finding vulnerabilities in critical infrastructure, accelerating biochemical research, producing manipulative content in election campaigns. A government that did nothing here would fail its protective duty.

On the other side, the manner of this specific order is troubling. It comes without public hearing. Without a transparent technical justification. Without a staged licensing scheme. And it lands on the very company that is locked in litigation with the same government over procurement access. Anyone who lays those three facts next to each other has earned the right to be sceptical. It is at the very least possible that security logic is being used as a tool to apply industrial-policy pressure. That possibility is a problem for any legal system that lives on trust in its own procedures.

I do not assess this in a single dimension. Reading the order as pure state arbitrariness would be naive. Reading it as routine security enforcement would be equally naive. It is both. And it sets a precedent that will reach far beyond Anthropic, because other providers will draw a lesson from this day: in a politically sensitive market, frontier providers have to expect a state-adjacent operating mode. The age of platform idyll is over.

The philosophical layer

Beneath the economic, legal, and political layers sits a philosophical question that gives the day its real weight. What is an AI model, actually? Is it a product that one sells? A service that one provides? An infrastructure that one operates? A form of knowledge that one shares? Or a collective capability that belongs to a society once it crosses a certain threshold of existence?

The question is not academic. The answer decides which legal shell fits. If a model is a product, contract law applies. If it is a service, consumer law applies. If it is infrastructure, network regulation applies. If it is knowledge, scientific freedom and academic fundamental rights enter the room. If it is a collective capability, an entirely different conversation begins, one about public goods, trusteeship, and ownership of socially relevant capabilities.

The June 12 order has answered this question without spelling out its answer. It has treated the model in practice as controlled dual-use technology, something between a good and a weapon. That is a strong categorisation. It carries consequences for how we talk in the years ahead about research, open weights, academic mobility, and international cooperation. It is not too early to lead that conversation openly. On the contrary. If we do not lead it, it will be led for us.

What companies should do now in practice

Anyone responsible for AI architecture in a mid-sized or large company has, since yesterday, a list of questions that does not wait. The list is not exhaustive. It is an order of priorities.

First, audit provider dependency

Which processes inside your organisation run today on a single frontier provider? Which of them are business-critical? Which of them could be migrated to a different model within 72 hours? If the third answer is not clear, you carry an open risk. You cannot cover that risk with contract clauses alone. You need technical substitutability.

Second, keep sovereign infrastructure as a live option

The conversation about sovereign cloud wrappers, EU data residency, and key custody has stopped being theoretical. It is now a procurement topic. If you build critical AI applications, you should have at least one documented option that can carry these applications even when the primary US model is no longer available. That is not political posturing. That is risk management.

Third, take open-weight strategies seriously

Open-weight models have made quality gains over the last 18 months that many have underestimated. They are not a frontier replacement for every task. But for a growing set of applications they are a genuine alternative. Anyone without a position on open weights should develop one. Ideally, with a concrete pilot use case.

Fourth, set a clean governance line

If the question of which model is used for which purpose lives on slides in your organisation but not inside a process, it belongs inside a process. That is not bureaucratic burden. It is the only way to survive in a market where providers can disappear overnight. If you have governance, you have options. If you do not, you live on luck.

The bigger arc

It is tempting to read the Anthropic export control of June 12, 2026 as an episode. A single order, a single company, a single conflict. That reading is too small. What happened in Washington that evening is the official end of a period in which frontier AI was treated as a global consumer good. That period ran roughly from 2022 to 2026. It produced a wave of democratisation that touched everyone from school children in Manila to law firms in Hamburg. And it built up an expectation structure that no longer holds in the new phase.

The next phase will look less spectacular. It will be shaped by licences, wrappers, audit reports, procurement clauses, and national stack strategies. It will be quieter, more formal, less consumer good, more utility. It will not be as friction-free as the previous phase. But it will rest on more realistic assumptions about how the world works, instead of on the assumption that tech stays tech and politics stays politics.

Anyone who wants to compete in this next phase has to think on three levels: technical substitutability, regulatory compatibility, and sovereign infrastructure options. If you ignore any one of those three, you will lose money. If you master all three, you will operate in a market where trust has become a scarce good again, and where trust is paid accordingly.

This is not the end of AI globalisation. It is the end of its innocence. And it is, for all the discomfort about the specific circumstances of this order, not a bad moment for an honest inventory.

Frequently asked questions

What exactly did the US Commerce Department order on June 12, 2026?

Commerce Secretary Howard Lutnick directed Anthropic to make Claude Fable 5 and Mythos 5 unavailable to any foreign national, both outside and inside the United States. Anthropic disabled both models globally for all users, because the existing infrastructure could not reliably differentiate by nationality. Other Claude models such as Opus 4.8 remain operational.

What legal basis does the order rest on?

The Commerce Department invokes the deemed export doctrine under 15 CFR 734.13 of the Export Administration Regulations. The norm was originally designed for semiconductors, encryption, and biotechnology and has never been applied to an AI model before. The AI Diffusion Rule of the Biden administration would have clarified the application to AI weights, but non-enforcement and its planned rescission were announced in May 2025.

Why were the models switched off globally instead of only for non-US users?

Anthropic states that the existing infrastructure does not allow clean filtering by nationality. In addition, a blanket exclusion of non-US citizens would have conflicted with the company's values and with US anti-discrimination law. A full shutdown was the fastest and legally cleanest reaction available.

Anthropic and the Pentagon dispute

What is the conflict between Anthropic and the Pentagon?

On February 27, 2026, the US Department of Defense designated Anthropic a „supply chain risk to national security". The background was Anthropic's refusal to remove contractual clauses prohibiting the use of its models for autonomous lethal weapons and mass domestic surveillance without specific cause. Anthropic filed two federal lawsuits on March 9, 2026, and opening arguments took place on May 19, 2026.

Is the export order connected to this dispute?

A direct causality has not been officially confirmed. The temporal and substantive proximity is, however, close enough that full independence is unlikely. The order targets the same company that is resisting certain government use conditions, and it cites a vulnerability that, according to Anthropic, is reproducible in comparable form across other frontier models.

Is Anthropic a victim or a provocateur in this conflict?

Both, in different proportions. Anthropic has positioned itself from the start as the ethically reflective AI company. That position is authentic self-binding and a commercial differentiator at the same time. The conflict with the Department of Defense rests on material substance and is at the same time media-cultivated. The June 12 order, on the state side, exceeds the bounds of a routine security action. There is legitimate criticism available in more than one direction.

Market impact

Which models remain available?

Anthropic continues to offer Claude Opus 4.8 and other earlier models. Other frontier providers such as OpenAI with GPT-5.5 or Chinese providers such as Qwen and DeepSeek are not affected by the order. In practical terms, many enterprise applications can be migrated, although the effort in prompt design and tooling layer is often significant.

Update, 4 October 2026: Fable 5 returned globally on 1 July; Mythos 5 initially returned for selected US organizations.

Do European companies have to change their AI strategy now?

The strategy itself does not need to be redrawn overnight. But any strategy that relies on a single US frontier provider is, as of yesterday, exposed to a concrete political risk. A technical migration path to a second provider, a documented open-weight option, and a sovereign infrastructure option should be in place over the coming months. This is risk management, not panic.

Do Chinese providers benefit from the order?

Very likely. Research labs and companies in Asia, Latin America, and the Middle East that had been using Mythos or Fable will port workflows to Qwen, DeepSeek, or comparable models over the coming weeks. Any block without a clear licensing logic accelerates substitution by accessible alternatives. This is not a geopolitical thesis, it is observable market rationality.

What does it mean for Mistral, Aleph Alpha, and Cohere?

European frontier providers gain an argument they could not have written themselves. Mistral CEO Arthur Mensch had already made the case for a sovereign alternative before the French parliament on May 17, 2026. The merger negotiations between Cohere and Aleph Alpha receive industrial-policy tailwind. The compute gap to the US providers remains large, however, and will not close within months.

Legal and political framing

Is the order lawful?

The question is open. The deemed export doctrine has existed for decades, but it has never been applied to an AI model. A clear statutory or regulatory basis specifically for AI model weights has not existed since the rescission of the AI Diffusion Rule in May 2025. On May 12, 2026, the Government Accountability Office held that the announced non-enforcement was a rule subject to the Congressional Review Act's submission requirements. A judicial review of this specific order is likely within months.

Is there a historical parallel in export-control history?

The closest parallel is the cryptography debate of the late nineties around the Wassenaar Arrangement. Strong encryption software was treated as dual-use technology. Cases such as Bernstein versus the United States and the work of Phil Zimmermann with PGP eventually drew a line between commercial standard technology and militarily sensitive specialist procedures. The current AI debate echoes the structure but operates at a much larger scale.

How is the European Union responding officially?

A formal EU response is not yet on the record as of this article. Early signals from Brussels suggest the order will feed into ongoing deliberations on procurement reform and cloud sovereignty criteria. France, Germany, and the Netherlands have flagged the risks of one-sided dependency in recent months. A coordinated European response is, however, likely to take weeks rather than days.

Strategy for companies and research

What should companies do in the next 30 days?

First, audit all business-critical AI processes for provider dependency. Second, build a documented migration path to a second frontier provider. Third, test an open-weight option for at least one productive use case. Fourth, set the AI governance line so that a model change can happen without organisational chaos. These four steps cost money, but they protect against considerably more expensive surprises.

What does the order mean for research and teaching?

Universities and research institutions with international staff have to examine which of their AI tools could fall under deemed export logic. This is especially relevant when US-based research partners share controlled tools with non-US researchers. The direct impact on teaching is limited, provided that alternative models are used. For international research cooperation, however, the order is a chilling signal.

Where can I find the most important primary sources?

The official statement from Anthropic is on the company's website. Reporting with primary source work appears at Fortune, Bloomberg, and Axios. The legal analysis of the GAO decision on the AI Diffusion Rule is available at JD Supra, among others. The background reporting on the Pentagon conflict is documented at GIS Reports, CSA Research, and CNBC.

Robotics and Automation in Practice: What EDAY 2026 in Vienna Revealed

There are moments when you look into a conference hall and realise a question has stopped being academic. The Julius Raab Saal of the Austrian Federal Economic Chamber, on the morning of 7 May 2026, was one of those moments. The rows were full. The first two seats in every row belonged to people with notebooks. People with coffee cups leaned against the back wall. The question hanging in the room was not whether robots are coming. They were already here. The question was who, in Europe, decides how they arrive.

The EDAY 2026, hosted by the Austrian Federal Economic Chamber (WKÖ), ran under the headline „Robotics and Automation - Investments with a Future". It is Austria's largest digitalisation event. This year, it took on a quality that was no longer just a conference. It was an early warning system.

The Number That Changes the Conversation

Eleven per cent. That is the share of Austrian companies that already deploy robots. In manufacturing, the figure rises to thirteen per cent. Among manufacturing companies with at least 75 employees, it climbs to forty-three per cent, and in those same companies roughly half of all processes are already automated. These are not trend statements from a consultancy slide. They are current results from a WKÖ survey, presented on the same day they were discussed, by people who produce such data for a living.

Read those numbers, then walk into an average industrial site in Austria or Germany, and the gap becomes visible immediately. The larger producers are tooling up. The small and mid-sized businesses stand in front of a door behind which someone is already at work, while they are still looking for the key. The honest task is not to open the door. The honest task is to make the key accessible without simplifying the things that genuinely are not simple.

Why Robotics Is Different This Time

For most of the past two decades, robotics in the public mind meant something heavy, yellow, fenced in, bolting parts to a car body in a line. That technology still exists, and it is mature and excellent. But the Raab Saal had something else on the table. Mobile robots crossing warehouse floors. Autonomous delivery vehicles from the Austrian postal service. Humanoid systems running test shifts next to humans in production. Robot cells that are no longer welded to the floor, but reconfigured in the morning because the product demands it.

The important shift is not that these systems exist. The important shift is that they increasingly share the same physical space with people. They are no longer fenced off. They are collaborative, mobile, learning. That changes the responsibility behind every single deployment. A robot cell is a technical question. An autonomous system sharing a walkway with humans is an organisational, legal and cultural one.

EDAY made the shift tangible. Gerald Greiner of BRP-Rotax presented humanoid robots in production. Dario Stojicic of ABB Robotics Austria talked through robot-assisted machine tending, with the real obstacles included. Kamran Karimi Moozarmi and Sebastian Blaha of the Austrian Post, together with Clarissa Groll of ACL, discussed automation and AI along the delivery process. None of these were visions. They were shift reports.

The European Decision Now on the Table

While operational reality moves forward, Europe wrestles with its place in the global robotics race. On one side stand volume and speed from the United States and China. On the other side stands a European claim that wants more than the fastest product. It wants a product that remains compatible with the society in which it is deployed. That is not a luxury position. It is an industrial condition.

The EU AI Act is in force and also covers AI in products and safety functions. For a physically acting machine, those requirements must be connected with machinery law, product liability and occupational safety. That is exactly the gap I have been trying to close with the concept of Robotic Governance since 2016. At EDAY, that gap stopped being a theoretical argument. It was a practical problem sitting between the entrepreneurs in the room and their advisors, insurers, and works councils.

The question I put to both halls was simple. If Europe does not decide, in the coming years, which robotics it wants, how to make it liable, how to teach it and how to use it, then the decision will be made for us. By the market. By suppliers. By standards from other economic zones. That is not a technological question. It is a question of sovereignty.

Keynote „Robotics, AI and the Next Industrial Decade - What Europe Must Decide Before Others Decide for Us" at EDAY 2026 in Vienna.

From Pilot Project to Production Line

At any conference of this kind, contributions divide into two categories. Some present demos. Others present shifts. Demos impress. Shifts convince. At EDAY, the share of shifts was unusually high. That is the genuinely noteworthy feature of this year.

Günter Renner of Internorm described how end-to-end automation links sales and production - one order, one continuous flow. Gerhard Anzinger of Anzinger Logistik delivered a sentence worth keeping: „Those who do not automate will lose. Those who automate badly will lose too." Hannes Watzinger of DigiTrans showed how automated driving creates concrete opportunities for Austria. Christoph Kandlhofer of voestalpine Signaling explained predictive plant ecosystems. Thomas Blumauer-Hießl of Siemens DAI drew the line back to the role of humans in autonomous systems.

The pattern underneath all these talks was almost the same. The technology is not the problem. The technology works. The problem sits one level lower and one level higher at the same time: in the quality of the data feeding the system, and in the clarity with which the organisation handles the exception case. Both are homework, not magic. Both routinely stay undone because they are less glamorous than the next demo.

Four Levers SMEs Can Pull Now

Sort through the day's contributions, filter out the buzzword noise, and four levers remain. They are realistic for small and mid-sized companies in the next twelve months. Not a recipe, more a sequence.

First: Name the Bottleneck Honestly

Automation pays where a specific bottleneck hurts: loading a machine, picking in the warehouse, a quality check that ties up half a Friday every week. Anyone who does not name the bottleneck first will buy technology against symptoms. That is more expensive than any consulting hour.

Second: Audit Your Own Data Without Flattering Yourself

Robotics without clean data is an expensive stage. If you have never duplicated a material number, mis-categorised a shift log or failed to record an exception case, you belong in a different book. Everyone else needs a quiet morning with their own data reality before a robot enters the building.

Third: Resolve Accountability Before Rollout

Who is allowed to stop the system when it formally works but produces nonsense in practice? Who is liable when the exception case arrives? Who decides when to retrain the system? Those three questions must be answered in writing before rollout. Not in a glossy slide. In a document that someone actually reads on a Monday morning.

Fourth: Use Existing Funding Programmes Pragmatically

The WKÖ made its instruments visible at EDAY: the AI Service Point, the KMU.DIGITAL funding scheme, the Innovation Map. Maria Lohmann from the RTR AI Service Point explained how SMEs handle the AI Act in practice. All of this exists today. Anyone who uses it saves weeks of solo research. Anyone who ignores it pays twice: once for the funding sitting unused in another pot, and once for their own consultant.

The Panel: An Honest Picture of Europe

After the keynote, Heidrun Bichler-Ripfel of the Institute for Applied Craft and Industry Research, Angelika Sery-Froschauer as Vice-President of the WKÖ, Dario Stojicic of ABB Robotics Austria, and Thomas Novak of the University of Applied Sciences Upper Austria joined me on the panel. The discussion had the rare tone that emerges when everyone around the table knows the subject and nobody is selling anything.

Three points stayed with me. First, Austria has an exceptionally capable industrial middle class that international observers underestimate. Second, the gap between sector champions and the SMEs that follow them will only close if the supply chain can think in continuous technological terms. Third, education is the underrated lever. The apprentice who walks into a training workshop this morning will, five years from now, operate systems that do not yet exist. That is not a threat. It is a requirement on curricula.

Opening discussion „Shaping the Future - Robotics and Automation in Practice" with Heidrun Bichler-Ripfel, Angelika Sery-Froschauer, Dario Stojicic, Thomas Novak and Dominik Bösl.

What Europe Must Decide Now

The central message of the day fits one sentence. Europe still has a window, but it is no longer wide open. The next industrial decade will be shaped by a small number of decisions taken in the coming three to five years.

Standardisation means that legal requirements and technical standards such as the new Machinery Regulation, ISO 10218 and the VDA 5050 interface recommendation for mobile robots are not only written, but lived on the shop floor. Sovereignty means that Europe develops its own platforms on which humanoid and mobile systems can be trained, without training data ending up on servers outside the continent. Education means that every technical school, every vocational academy, every university of applied sciences audits its curricula against a future in which autonomous mobile systems are part of the inventory.

And liability means closing the gap between the EU AI Act, the Machinery Regulation and product liability law, before courts have to decide cases without a clear political line. That is the most uncomfortable of all options.

EDAY Through the Press

The press coverage the day after EDAY largely matched the picture inside the hall. The official WKÖ press release on OTS framed the event as a „growth driver for Austrian businesses" and highlighted the robotics survey data. OE24.tv ran a television feature on „Austria's largest digitalisation event". The Austrian startup outlet Brutkasten framed the day under the question of whether robotics and automation can be drivers of future growth - a question the room itself had already answered. TOP News Österreich and elektro.at added further reporting.

What Stays

A good conference day is not the one that sends you home enthusiastic. A good conference day is the one that rewrites the next three weeks of the agenda. EDAY 2026 managed that because three things happened at the same time. Reliable numbers were on the table. The right practitioners were in the room. And a political question was raised that cannot be delegated.

The question is who shapes robotics in Europe in the coming years: we, or someone else. If I had to place a bet after this day, it would be this one. Austria and the wider German-speaking region carry more substance, more talent and more industrial experience than the international perception suggests. That is not enough on its own. It still takes someone to decide, before rollout, what happens when the system fails. At the level of a single factory hall. And at the level of a continent.

As long as that question remains open, eleven per cent of robot adoption is a pleasant number. Once it is answered, those numbers become the structural foundation of the coming industrial decade.

Frequently Asked Questions about Robotics, Automation and EDAY 2026

What was the theme of EDAY 2026?

EDAY 2026, hosted by the Austrian Federal Economic Chamber (WKÖ), ran under the headline „Robotics and Automation - Investments with a Future". The focus areas included robotics in manufacturing, AI-based applications for SMEs, digital sovereignty, cybersecurity and the practical handling of the EU AI Act in Austrian businesses.

How many Austrian companies already use robots?

According to the WKÖ survey presented at EDAY 2026, 11 per cent of Austrian companies already deploy robots. In manufacturing the share is 13 per cent, while manufacturing companies with at least 75 employees reach 43 per cent. In these larger companies, around 50 per cent of processes are already automated.

What does Robotic Governance mean in practice for SMEs?

Robotic Governance is the regulatory and ethical framework for autonomous, physically acting systems. For SMEs it boils down to three priorities: clear accountability before rollout (who can stop the system, who is liable in the exception case), alignment with applicable machinery law and relevant technical standards such as ISO 10218, with VDA 5050 as an interface recommendation for mobile robots, and a realistic approach to the EU AI Act and the new Machinery Regulation.

Which levers should SMEs use right now for robotics and automation?

Four levers are realistic and feasible within twelve months: first, name the concrete bottleneck instead of buying technology against symptoms. Second, audit your own data honestly before placing robotics on top. Third, resolve accountability and liability in writing before rollout. Fourth, use existing funding programmes such as KMU.DIGITAL, the AI Service Point at RTR and the WKÖ instruments pragmatically.

Why is the EU AI Act not enough for robotics?

The EU AI Act also covers AI in products and safety functions. A complete robotics project additionally needs machinery law, product liability, technical standards and operational responsibility. Robotic Governance connects these levels with ethics, incentives and societal dialogue.

Where can I watch the EDAY keynote and panel discussion?

Both recordings are available in the official WKÖ YouTube playlist. The keynote „Robotics, AI and the Next Industrial Decade" is at youtube.com/watch?v=u7csYc6a_iY, the opening panel „Shaping the Future - Robotics and Automation in Practice" at youtube.com/watch?v=qHiOE_TRiKQ. Both videos are part of the official EDAY 2026 playlist on the WKÖ channel.

Which European decisions are critical for the next industrial decade?

Four decisions are time-sensitive: standardisation (European norms for humanoid and mobile robots applied on the shop floor), sovereignty (European platforms for training autonomous systems without exporting training data), education (curricula in vocational schools, universities and applied science institutions that anticipate autonomous robotics) and liability (closing the gap between EU AI Act, Machinery Regulation and product liability law before courts have to decide without clear political guidance).

Further reading, October 29, 2026: My article on AI and the future of work examines how organizations can connect technology, work design and competence development, with research, a workflow example and practical steps.