President Emmanuel Macron used the G-7 summit in Évian-les-Bains on Wednesday to press allied governments and American executives on a “trusted partners” framework that would restore access to two Anthropic models the Trump administration ordered offline four days earlier. The diplomatic choreography, more than the communiqué itself, was the point: Europe is now negotiating with Washington over the terms on which it gets to use American software at all.

The trigger was a June 13 export-control directive instructing Anthropic to suspend foreign-national access to Fable 5 and Mythos 5. Because American “deemed export” rules treat any access by a foreign national, including a company’s own employees, as an export, Anthropic was forced to pull both models offline worldwide. Fortune, writing on the eve of the summit, described the resulting regime as a de facto licensing system for frontier A.I., assembled ad hoc after researchers at Amazon found a way to jailbreak some of Fable’s cybersecurity guardrails.

Macron met on the margins with Anthropic’s Dario Amodei and OpenAI’s Sam Altman, with further bilaterals planned involving Meta Platforms and OpenAI executives. According to a diplomat familiar with his thinking, cited by Bloomberg, the French president sees a trusted-partners scheme as the only near-term route to restoring allied access without asking Washington to reverse its security finding. Macron said he expected progress “in coming weeks” on broadening access to Anthropic’s Mythos system, and argued that nobody would buy American A.I. if customers feared it could be switched off without warning.

The working lunch surfaced the divergence. Amodei argued that the breakthroughs, though led by American firms, drew on researchers from each of the G-7 nations, and that the benefits should be shared among democracies. Altman, seated beside President Trump, told governments: “Do not cede your responsibilities to A.I. labs like mine. We develop the technology, and the citizens of the free world make the rules.” Trump, per a person familiar with the meeting cited by The Washington Post, didn’t address Anthropic or the ban; he talked about staying ahead of China and easing approvals for new electricity generation.

Demis Hassabis of Google DeepMind also attended. The G-7 communiqué tasked finance officials, regulators and cybersecurity experts with assessing how frontier models could affect financial stability, productivity and labor markets, the same instruments Basel reached for after 2008. The recognizable shape, again: a security tool built unilaterally in Washington, an allied scramble to legibilize it, and the labs themselves volunteering to be governed.

Sources