The final day of the G7 summit in Évian-les-Bains produced an extraordinary tableau: the leaders of the world’s richest democracies sitting down to a working lunch with the chief executives of OpenAI, Anthropic, Google DeepMind and the French start-up Mistral. Convened by President Emmanuel Macron, the meeting was billed as a session on securing artificial intelligence, but it was immediately overshadowed by a unilateral American decision. Days earlier, Washington had ordered Anthropic to block access for all non-US citizens to its two newest models, Fable 5 and Mythos 5, citing national security. The company, unable to filter users by nationality, simply deactivated the models for everyone. That abrupt move turned the Évian lunch into an impromptu crisis summit on the geopolitics of AI.

Viewed from Paris and Berlin, the Anthropic affair is a sharp wake-up call. Macron used the gathering to float the idea of a “trusted partners” framework that would allow allied nations to access advanced American models under agreed safeguards, effectively seeking a diplomatic bypass around Washington’s export controls. The German chancellor, Friedrich Merz, reinforced the message, arguing that the potential of new technologies “must be available to all countries” and that the episode showed Europe “needs to update itself.” The G7 communiqué promised a new “cooperation platform” to manage AI risks and develop common standards, with a follow-up meeting planned for September. Behind the scenes, the European Union’s cybersecurity agency separately arranged talks with Anthropic in San Francisco, signalling that Brussels intends to pursue both multilateral and bilateral channels.

From Washington’s perspective, the calculus is starkly different. President Donald Trump, who attended the lunch and later dined at Versailles, told journalists that negotiations with Anthropic were “going well,” offering no detail but a characteristically transactional note of optimism. Yet the US government has now effectively blacklisted Anthropic twice, a pattern that Fortune reported is complicating the company’s plans for a blockbuster initial public offering later this year. Investors are being forced to weigh whether a trillion-dollar valuation can hold if Washington can shutter flagship products overnight. American officials insist they can work with allies on minimising AI’s security risks, but the export ban has been met with accusations of hypocrisy in parts of the European press, which note that the G7’s rhetoric about an open and safe digital space sits uneasily alongside a unilateral embargo on the very technology under discussion.

The Évian lunch may be remembered less for its communiqué than for the fault line it exposed. The G7’s proposed platform is a tentative step toward multilateral governance, yet the Anthropic case demonstrates that the most consequential decisions about frontier AI are still taken in national capitals, not international summits. For Europe, the episode has lent urgency to calls for strategic autonomy, embodied by the presence of Mistral’s Arthur Mensch at the table. Whether the September meeting can begin to bridge the gap between Washington’s security imperatives and the rest of the world’s demand for access will determine if AI becomes a domain of shared rules or an instrument of technological fragmentation.