European leaders pressed Dario Amodei, Demis Hassabis and Sam Altman at a June 17 G7 working lunch in Evian-les-Bains over whether governments and businesses can rely on U.S.-built frontier AI after a U.S. Commerce Department directive led Anthropic to cut off its Fable 5 and Mythos 5 models worldwide, according to the source account. The fallout matters because European public bodies and companies had already built those systems into day-to-day operations.

The lunch, hosted by French President Emmanuel Macron during the June 15-17 summit, was devoted entirely to artificial intelligence. The source material says Amodei of Anthropic, Hassabis of Google DeepMind and Altman of OpenAI sat with heads of state and senior officials, including Macron, German Chancellor Friedrich Merz, European Commission President Ursula von der Leyen, U.K. Prime Minister Keir Starmer and U.S. President Donald Trump.

The immediate trigger was the June 12 U.S. export-control directive. According to the source account, it ordered Anthropic to block its most capable models for any ‘foreign national’. Because real-time nationality checks are difficult at API scale, the account says Anthropic responded with a global cutoff, leaving European users without lead time or a phase-in period.

The AI executives offered a governance track rather than a direct fix. Amodei called for a U.S.-led coalition of democratic states, Hassabis backed a Western alliance, and Altman proposed an international forum for testing standards. European officials, by contrast, pressed for reliable access, limits on future cutoff risk, trusted-partner rights, domestic compute capacity and child-safety rules built into AI systems.

Europe Wants Access Guarantees

The meeting turned AI dependence from a policy concern into an operational risk. If a foreign export-control decision can remove access to models used in business, public services or research, then procurement decisions involving frontier AI carry continuity risks beyond ordinary vendor failure.

For Europe, the issue is also industrial policy. Macron’s camp is pushing for domestic capacity, including compute sites, power, chips and European or allied model providers, while still seeking access to the best U.S. systems. That balance matters for companies deciding where to build AI products and for governments writing rules for sensitive sectors.

June Shutdown Reframed The Summit

The source account says the U.S. directive arrived five days before the lunch and remained in place after the summit. Invited tech leaders included Salesforce’s Marc Benioff, Meta AI chief Alexandr Wang and leaders from Mistral, Synthesia, Black Forest Labs, Domyn and Sakana AI, signaling that the discussion was not limited to three U.S. firms.

Macron’s wider answer was a sovereignty push: a EUR 420 billion package, AI gigafactories and the CADA initiative cited in the source material. The G7’s public theme was ‘safe, rapid, and effective deployment of AI’, but the private stakes described in the source were control, access and infrastructure location.

“Democracies should not give in to the temptation to splinter.”

— Dario Amodei, Anthropic chief executive

Guarantees Remain Outside CEO Control

It is not yet clear whether Washington will narrow, pause or reverse the Anthropic directive. The source account says the ban stayed in place after the summit, and the three executives could not by themselves promise that a future U.S. order would not override commercial contracts.

Key design questions are unresolved: which countries would qualify as trusted partners, whether access rights would have legal force, how China-related chip limits would be policed, and how child-safety principles would be written into products used across different legal systems.

September Talks Test The Proposal

The source material says leaders plan to build a platform for Western democracies within a month and return to the issue at a September meeting. Expected agenda items include trusted-partner access, cyber-defense cooperation, shared testing standards, AI infrastructure planning and common principles for child and youth safety.

The next milestone will show whether Europe receives enforceable access commitments or mainly new forums for coordination. Until then, European governments and companies remain exposed to the same basic risk raised by the Anthropic cutoff: core AI services can be affected by decisions made outside their own jurisdictions.

Key Questions

What happened at the G7 AI lunch?

Macron hosted an AI-focused working lunch on June 17 with G7 leaders, senior officials and AI executives including Amodei, Hassabis and Altman. According to the source material, the meeting centered on the fallout from Anthropic’s sudden model cutoff and Europe’s demand for reliable access.

Why did Anthropic cut off Fable 5 and Mythos 5?

The source account says a June 12 U.S. Commerce Department export-control directive required Anthropic to block those models for any ‘foreign national’. Because the company could not reliably verify nationality in real time at API scale, it applied a worldwide block.

What does Europe want from the AI companies?

European leaders want durable access to frontier models, protection from another sudden cutoff, a trusted-partner scheme for non-U.S. allies, a role in infrastructure decisions and stronger child-safety commitments. They are also backing European compute and model capacity.

Can Amodei, Hassabis and Altman provide those guarantees?

Only partly. They can support standards, access programs and safety forums, but export controls are set by governments. The source material’s central tension is that the executives do not control the switch Europe is worried about.

When will this be revisited?

The source account says leaders are aiming for a platform within a month and another meeting in September, when trusted-partner access and safety standards are expected to return to the agenda.

Source: Thorsten Meyer AI