AI leaders joined G7 discussions in Evian-les-Bains, France, on June 17, 2026, as governments debated how to deploy artificial intelligence while reducing dependence on U.S.-controlled frontier models. AP reported that OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, and Anthropic CEO Dario Amodei were due at a working lunch on safe, rapid, and effective AI deployment, alongside leaders from Cohere, Mistral, Black Forest Labs, Domyn, Sakana AI, and Synthesia.
The timing matters because the meeting followed Anthropic’s Fable 5 and Mythos 5 suspension. Anthropic said on June 12 that a U.S. government export-control directive forced it to disable those models for all customers to ensure compliance with a restriction on foreign-national access.
Access is no longer a procurement detail
For most enterprise buyers, model access has looked like a vendor decision: choose the provider, negotiate the contract, set the data terms, and build. The Anthropic episode made the dependency more visible. If a government can abruptly limit access to a model class, then the buyer’s risk is not only uptime or price. It is jurisdiction.
AP framed the G7 discussion around European concerns over American dominance and tech sovereignty. That is not abstract. Anthropic said the directive applied to foreign nationals whether inside or outside the United States, including foreign-national Anthropic employees. The company also said access to all other Anthropic models was unaffected, but the targeted suspension still exposed how a frontier model can become controlled infrastructure.
The G7 table was broader than the big three
The attendee list matters. AP said the working lunch included leaders from U.S. frontier labs and smaller or non-U.S. AI companies, including Canada’s Cohere, France’s Mistral, Germany’s Black Forest Labs, Italy’s Domyn, Japan’s Sakana AI, and U.K.-based Synthesia.
That mix reflects the policy tension. The largest models and platforms are still heavily U.S.-centered. But governments want local compute, local data control, local industrial capacity, and at least some domestic or allied model options. Cohere CEO Aidan Gomez told AP the company’s focus at the G7 was expanding sovereign AI ecosystem partnerships beyond Canada and Germany.
The European Council’s June 17 statement package from Evian listed emerging technologies among the topics discussed, but the published joint statements did not amount to a single binding AI-access agreement. That distinction matters. The story is the agenda shift and the dependency exposed by the Anthropic directive, not a finished G7 compact.
Anthropic became the case study
Anthropic’s own statement is unusually direct. The company said the U.S. government cited national security authorities and did not provide specific details of its concern. Anthropic said its understanding was that the government had seen a method of bypassing Fable 5, but the company argued the finding did not justify recalling a commercial model deployed to customers.
That disagreement is the policy fault line. Governments want control over powerful cyber and bio capabilities. Model companies want predictable release paths and customer trust. Allies want assurance that they will not be cut off by U.S. decisions made under national-security pressure.
The counter-case is that not every access restriction is unreasonable. Frontier models can have capabilities that create real misuse risk. A trusted-access program may be the right tool for some domains. But the more important the model becomes, the more every restriction looks like infrastructure policy.
What buyers should take from this
Enterprise and government teams should start treating frontier-model access as a resilience question. That means asking which workloads can move across providers, which depend on a specific model’s capability, where data and inference are hosted, and what happens if a model tier is suspended or region-limited.
For regulated industries, the answer may be a portfolio: hosted frontier models for the hardest work, smaller local models for continuity, and contractual clarity around notice, export-control exposure, and incident response. For governments, the answer may include domestic model efforts, compute investments, and allied access agreements.
The G7 discussion did not settle that architecture. It did make the question harder to avoid. Model capability is now part of national economic strategy, and access can be changed by law, not only by product roadmap.
For readers tracking the companies and models behind these policy fights, see our AI model leaderboard and AI company tracker.
