Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis presented a formal proposal for a US-led international AI coalition at the G7 summit in Evian-les-Bains, France on June 17, 2026. The pitch: structured access to frontier models for allied governments, coordinated trade policy on chips and components that deliberately excludes China, and joint work on AI risks in cyber operations, bioterrorism, and intelligence analysis.
The lunch drew roughly a dozen technology executives alongside G7 heads of state including President Donald Trump, Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and Secretary of State Marco Rubio. OpenAI’s Sam Altman attended alongside executives from Mistral, Cohere, Black Forest Labs, Synthesia, Salesforce, and Meta. Altman floated a companion idea: an international forum to establish globally accepted standards for model testing and to produce impartial analysis of frontier capabilities and risks. Canadian Prime Minister Mark Carney, according to people familiar with the discussions, agreed that the United States should lead such a coalition.
The timing carries a specific irony. Five days before the summit, on June 12, the Trump administration placed export controls on Anthropic’s own Fable 5 and Mythos 5 models, citing national security concerns following reports of a jailbreak vulnerability. The controls blocked all non-US citizens from using the models, including Anthropic’s own non-citizen employees. Anthropic disabled both models worldwide while it negotiated with Commerce Department officials. The summit marked the first face-to-face meeting between Amodei and Trump since those restrictions took effect.
The proposal’s core vocabulary is worth unpacking. “Structured access” in the context of frontier models typically means tiered permissions, use-case restrictions, and monitoring requirements baked into the access agreement. Applied across allied nations, it would formalize something the industry has so far handled informally: who gets to run the most capable models, for what purposes, and under whose oversight. The China exclusion on chips and components is not new policy. It extends existing Bureau of Industry and Security controls into a multilateral coordination framework, which would be harder for individual allies to quietly route around.
The gathering was explicitly conversational. The Next Web and CNBC both reported that it produced no binding commitments and no regulatory announcements. It put the governance question in front of heads of state and gave two of the largest frontier labs a direct channel to frame the terms of that debate. The absence of formal output is also informative: the administration that just export-controlled Anthropic’s own models is the same administration being asked to lead a coalition on behalf of those models’ creators.
For builders, the relevant signal is not what was announced at the summit but what the two proposals together imply. Amodei’s coalition framework and Altman’s standards forum both assume that frontier model access will become a managed resource: allocated by governments, tested against shared benchmarks, and conditioned on security and use-case compliance. If that framework takes shape, teams building on frontier APIs outside the US will face a compliance layer that does not currently exist.
Any non-US team currently planning multi-year products on frontier model access should treat this proposal as a leading indicator that the terms of that access are subject to renegotiation at the state level, not just the vendor level.
Reporting by The Next Web and CNBC, published June 17, 2026.
