The EU's AI plan: autonomous driving tests, health data and farm advisers by December
The European Commission appears to be changing course in its AI policy: rather than chasing the US and China in the race to build the leading language models, Europe should focus on deploying artificial intelligence faster and more widely across the economy. The message comes from Jim Hagemann Snabe, the Commission's own AI envoy, according to Politico's account of internal documents — relayed through heise online, The Next Web and The New York Times. None of this is official EU policy yet: the strategy rests on unverified internal notes and unnamed officials, and the Commission has not decided anything final.
The message from 4 September
The core of the reported shift was presented to all 27 commissioners at a private seminar on 4 September, held at a palace outside Brussels. Two Commission officials have told Politico that Snabe delivered the message there; The New York Times has confirmed the meeting through two people briefed on the gathering, writing that Snabe outlined ways Europe could use the technology for economic gain, while others raised concerns about AI safety and increasing risk (NYT).
The presentations Snabe's team used — described by TNW, citing Politico — carried a pragmatic main point: the US and China may be ahead on models and computing power, but Europe can still win by adopting AI. The contest is thus not over who builds the best frontier model, but over who converts the technology into productivity fastest in industry, healthcare and public administration.
The concrete programme
Internal documents seen by Politico give three examples of what an "application-first" strategy could mean in practice:
- Easier rules for testing autonomous vehicles — reducing friction for industrial pilot projects.
- Secure data rooms for personalised medicine — infrastructure that makes sensitive health data available for AI development without opening it up uncontrollably.
- A digital adviser for farmers — AI assistants in agriculture.
The timeline is concrete too: by December, the Commission is to assess the use of AI across economic sectors. Snabe's team is going through pilot projects and infrastructure to present a package of measures by the end of the year, and a final strategy plan is to follow in early 2027 (heise).
Who is Snabe?
Jim Hagemann Snabe is chair of Siemens and has been appointed as AI adviser to Commission President Ursula von der Leyen and technology chief Henna Virkkunen. The sources disagree on the timing: TNW writes that he was appointed in June, heise says July — the difference cannot be resolved with the available documentation. The appointment drew criticism at the time over conflicts of interest, since he simultaneously chairs one of Europe's largest industrial groups.
Even with the new focus on use, Snabe warns against dependence on foreign suppliers. According to heise, Europe must ensure its own companies can develop, test and scale their ideas, and a significant share of value creation must remain in Europe. The push for application is thus not the same as giving up the industrial base — but how the two ambitions are to be reconciled remains unresolved.
Von der Leyen's balancing act
The Commission President has herself adopted a formulation that accommodates both growth and caution. In her State of the Union speech on 16 September she said: "AI comes with a balance of opportunities and risks", adding: "Without addressing these risks, we will never be able to unlock AI's potential" (NYT).
The NYT reports at the same time that the Commission has not reached any final conclusions since the 4 September meeting, and that several discussions between European leaders are planned over the coming year. According to the newspaper, von der Leyen and the commissioners still have to work out what they can do about a technology that is changing rapidly. In other words: there is no adopted strategy yet, only a course under discussion.
The opposition: enforcement and computing power
The shift is contested inside the EU before it is even final. Two alternative lines emerge in the source material:
Axel Voss and enforcement. The German Member of the European Parliament has, according to Politico as relayed by TNW, criticised the Commission for enforcing the AI Act too slowly, and pointed, according to the same source, to recent hacker attacks carried out by AI agents. Voss's point is fundamentally different from Snabe's: the problem is not that Europe is betting on the wrong thing, but that the existing rulebook is not being enforced fast enough against real risks.
Vestager and Aghion on computing power. Former Commissioner Margrethe Vestager and the economist Philippe Aghion have during September called for Europe to control 15 percent of global computing power by 2030. That is a far more capital-intensive strategy than Snabe's application focus — a direct alternative to the idea that Europe can get by by facilitating the use of other people's models.
These positions are reported as rival viewpoints, not adopted lines. What they share is that they point to the same tension: how much should the EU invest in its own resources (models, compute, safety), and how much can be achieved by enabling the use of existing technology?
Why it matters
If the pivot becomes official policy, it would represent a significant change in the EU's AI profile. Since the AI Act was passed, Brussels has largely been defined by regulation. A strategy that uses expertise and rules to lower the threshold for deployment — in transport, healthcare and agriculture — would shift the emphasis from restricting technology to distributing it. For European industry and the public sector, it could mean faster access to pilot schemes and data-sharing infrastructure.
At the same time, the entire picture rests on sources: Politico's reading of internal documents and slides, unnamed Commission officials, and people briefed on closed meetings. Neither the slides nor the package of measures has been independently confirmed, and the Commission has published no final conclusions.
What to watch
Three dates will determine whether the shift becomes reality. December will show whether the Commission actually delivers the promised sector assessment and package of measures. The final strategy plan is due in early 2027. And in the meantime, the internal counter-positions — Voss's demand for faster enforcement of the AI Act, and Vestager and Aghion's demand for European computing power — will determine how much of the application focus survives in the compromises. Until then, the accurate picture is a course taking shape, not one that has been decided.