The European Commission is in direct talks with OpenAI and Anthropic over their frontier AI models, according to Reuters. The engagement puts Brussels in conversation with the two US labs that sit behind ChatGPT and Claude, the dominant Western general-purpose AI systems.
The Commission has not publicly detailed the scope of the discussions. Neither OpenAI nor Anthropic has issued a statement describing what is on the table. What is clear is that the EU executive wants a working channel with the model builders themselves, not just the cloud platforms and downstream deployers that resell their systems inside the bloc.
The talks land as the EU works through the implementation phase of its AI Act, the bloc's horizontal AI law. The Act sets out a separate regime for general-purpose AI models, with heavier obligations on systems judged to present systemic risk — a category that, on any reasonable reading, captures the flagship models from OpenAI and Anthropic.
Key facts
- 01The EU Commission is in direct talks with OpenAI and Anthropic over their frontier AI models, according to Reuters.
- 02The discussions involve the two US labs behind ChatGPT and Claude, the dominant Western frontier model providers.
- 03Neither the Commission nor the companies have detailed the scope of the talks publicly.
Those obligations include model evaluations, adversarial testing, incident reporting, cybersecurity protections and transparency on training data summaries. For the labs, the practical question is what counts as compliance evidence: which evaluations, run by whom, against which thresholds. Direct dialogue with the Commission is how those operational definitions get pinned down before enforcement, rather than litigated after it.
“Brussels wants a direct line to the two labs setting the frontier — a sign the AI Act's enforcement phase will be negotiated with model builders, not just dictated to them.”— Jaeden Schafer
OpenAI and Anthropic have both spent the past two years building out European-facing policy teams. OpenAI is the larger commercial presence, with ChatGPT consumer and enterprise products live across EU member states. Anthropic, smaller but growing fast, sells Claude through its own API and through Amazon Web Services and Google Cloud, both of which serve European customers from in-region data centres.
Brussels has been explicit that it views frontier models as a strategic category. The Commission set up an AI Office inside DG CNECT specifically to supervise general-purpose models, and that office is the most likely Commission interlocutor in any technical conversation with the two labs. Its mandate covers evaluations, codes of practice, and the day-to-day relationship with model providers placing systems on the EU market.
For the labs, the calculus is straightforward. The EU is a large, regulated market they cannot afford to exit, and the cost of getting on the wrong side of the AI Office — through delayed product launches or restricted deployment of new model versions — is higher than the cost of detailed engagement. OpenAI delayed European launches of some ChatGPT features in prior years over data-protection questions; neither company wants a repeat under the AI Act.
There is also a competitive overlay. European policymakers have repeatedly said they want a domestic frontier-model industry, and have pointed to Mistral and a handful of national champions as evidence one is emerging. Direct talks with the two leading US labs cut against any pretence that the bloc can regulate frontier AI without engaging the companies that actually build it.
The structure of these conversations matters for the rest of the industry too. Whatever evaluation protocols, reporting templates and red-team disclosures OpenAI and Anthropic agree to with the Commission will set the de facto template that smaller model providers — European and otherwise — have to match. Compliance regimes tend to harden around the first negotiated cases.
What the sources do not yet establish is whether the talks are exploratory or tied to a specific enforcement question, a specific model release, or the drafting of the AI Act's code of practice for general-purpose models. Until the Commission or the companies say more, the substance of the discussions remains a black box.
The signal worth tracking is direction of travel. The EU is no longer treating frontier model regulation as something that happens to the AI industry from the outside; it is sitting across the table from the two labs that define the frontier. For OpenAI and Anthropic, that is both a constraint and an opportunity — every clause they help shape now is a clause they will not be litigating later, and every concession Brussels extracts becomes the floor for everyone else selling a general-purpose model into Europe.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
