TL;DR
European leaders used a June 17 G7 working lunch with Dario Amodei, Demis Hassabis and Sam Altman to press for durable access to frontier AI models after a U.S. export-control directive forced Anthropic to block Fable 5 and Mythos 5 for foreign nationals. The confirmed result so far is pressure for a trusted-partner framework, common testing standards and European AI capacity, not a reversal of the U.S. restriction.
European leaders used a June 17 G7 working lunch in Évian-les-Bains to press the heads of Anthropic, Google DeepMind and OpenAI for dependable access to frontier AI systems after a U.S. export-control directive forced Anthropic to block its Fable 5 and Mythos 5 models for foreign nationals, disrupting European users and exposing how quickly foreign-built AI infrastructure can be withdrawn.
The lunch brought Dario Amodei, Demis Hassabis and Sam Altman into the same room as heads of state and senior ministers during the June 15-17 summit. Reporting by Business Insider, the Financial Times and Axios said the session also included other tech leaders, among them Salesforce’s Marc Benioff, Meta’s Alexandr Wang and executives from Mistral, Synthesia, Black Forest Labs, Domyn and Sakana AI.
The immediate trigger was the June 12 Commerce Department directive described in summit reporting. The order required Anthropic to block access for foreign nationals to its most capable models, Fable 5 and Mythos 5. Because real-time nationality checks are difficult at API scale, Anthropic imposed a wider shutdown for affected access, according to the summit account. European businesses and public bodies that had built those models into workflows lost access without advance notice.
At the table, the three executives offered variations on coordinated governance rather than unilateral company control. Amodei called for a U.S.-led coalition of democratic states with structured access for trusted partners and chip trade that excludes China. Hassabis backed a Western coalition. Altman proposed an international forum for testing standards and risk analysis. European leaders, by contrast, pressed for durable access rights, protection against another abrupt shutdown, a trusted-partner route for non-U.S. allies, more say over compute and chip infrastructure, and stronger child and youth safety rules.
Évian and the fallout: what Europe actually wants
For the first time, Amodei, Hassabis, and Altman sat with heads of state — five days after Washington switched Anthropic’s models off worldwide. Europe’s question: can you rely on models a foreign cabinet can shut down by decree?
The dilemma: what Europe wants from the three CEOs, the three can’t deliver — because they don’t hold the switch, Washington does. Macron’s platform is the right answer, but no fix for a decade-old infrastructure gap. The only answer that doesn’t depend on someone else’s goodwill: your own models, your own compute, open weights you can self-host.
Europe Wants More Than Access
The dispute matters because many companies, public agencies and research groups are building operations around models they do not own, cannot host and cannot fully control. The Anthropic shutdown turned a policy debate about dependency into an operational risk: a decision in Washington can reach European hospitals, banks, schools, software vendors and government services if those organizations rely on U.S.-hosted frontier models.
The episode also changes the negotiating position of the AI labs. Amodei, Hassabis and Altman can offer standards bodies, safety forums and partner programs, but they cannot promise that U.S. export controls will not interrupt access again. That gap is why European officials are linking AI safety to sovereignty: model access, data-center siting, energy supply, chip allocation and local hosting are now treated as pieces of the same policy problem.
AI model access control API
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The June 12 Shock
The Évian lunch followed years of G7 work on AI governance, including the Hiroshima AI Process and later safety summits, but the June 12 directive gave the discussion a sharper edge. The public theme for the lunch was safe, rapid and effective AI deployment. The private pressure point, according to summit reporting, was whether allied users can depend on U.S. models when U.S. national-security officials can restrict them with little warning.
The European response also fits a broader industrial push. Summit reporting says Europe is tying the access dispute to a planned €420 billion technology-sovereignty package, AI gigafactories and the CADA initiative. Those plans are meant to reduce reliance on foreign model providers and foreign compute, though building competitive capacity will take years.
“resist the temptation to splinter”
— Dario Amodei, Anthropic CEO
enterprise AI safety testing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Guarantees Still Lack Teeth
No public agreement has yet shown how a trusted-partner system would work, who would qualify, or whether access rights could survive a later U.S. security order. It is also unclear how long the Anthropic restrictions will remain in force, what conditions Washington would require for restoration, and how many European users were affected.
Several points remain claims rather than confirmed outcomes. AI executives say democratic coordination can preserve access while managing risk; European leaders say local compute and homegrown models are needed to reduce exposure. Neither side has yet produced a binding mechanism that resolves the conflict between alliance access and U.S. export-control authority.
trusted partner AI infrastructure
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Platform Talks Move to September
According to summit reporting, Macron’s AI platform is expected within a month, and leaders are due to reconvene in September to discuss democratic AI cooperation, trusted-partner access, cyber-defense coordination and child-safety principles. No reversal of the Anthropic restriction has been confirmed.
The next test is whether the follow-up process produces more than statements: a clear access regime for allies, shared testing rules, safeguards for minors and credible European investment in models and compute. Until then, frontier-model access remains a policy risk as well as a technical choice.
AI model security software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened at the G7 AI lunch in Évian?
G7 leaders met with senior AI executives, including Dario Amodei, Demis Hassabis and Sam Altman, on June 17, 2026, to discuss AI deployment, safety and access after the U.S. restriction on Anthropic models disrupted foreign users.
Why did Anthropic restrict Fable 5 and Mythos 5?
According to the source account, a June 12 U.S. Commerce Department directive required Anthropic to block the models for foreign nationals. The U.S. cited national-security concerns, but the exact review process has not been fully made public.
Did Europe get the Anthropic ban reversed?
No. The summit produced pressure for follow-up talks, trusted-partner access and shared standards, but no confirmed reversal of the restriction.
What does Europe want from the AI chiefs?
European leaders want reliable access to frontier models, protection against sudden shutdowns, a trusted-partner scheme for allies, influence over compute infrastructure, more European AI capacity and stronger safety rules for children and young people.
Why are self-hosted models part of the debate?
Self-hosted or open-weight models can reduce dependence on a foreign provider’s API and on another government’s export-control decisions. They do not remove every risk, but they give users more control over continuity.
Source: Thorsten Meyer AI
