TL;DR
European leaders used the G7 AI lunch in Évian-les-Bains to demand more reliable access to advanced AI after a reported U.S. export-control order led Anthropic to shut off Fable 5 and Mythos 5 worldwide. The companies offered cooperation on standards, security and trusted access, but the central problem remains political: the U.S. government, not the CEOs, controls export orders.
European leaders used the G7 summit in Évian-les-Bains, France, to press the heads of Anthropic, Google DeepMind and OpenAI over reliable access to advanced AI after a reported U.S. export-control order led Anthropic to cut off its newest models worldwide, turning a policy dispute into an immediate business and public-sector risk.
The AI-focused working lunch on June 17 put Dario Amodei, Demis Hassabis and Sam Altman at a table with G7 leaders and senior officials, including Emmanuel Macron, Friedrich Merz, Ursula von der Leyen and Keir Starmer. Other tech leaders and allied AI firms were also invited, including Salesforce’s Marc Benioff, Meta AI chief Alexandr Wang, France’s Mistral, Britain’s Synthesia, Germany’s Black Forest Labs, Italy’s Domyn and Japan’s Sakana AI, according to the source material and public reporting from the Financial Times and Business Insider.
The immediate trigger was a June 12 U.S. Commerce Department directive that, according to reports cited in the source material, ordered Anthropic to block access to its most capable models, Fable 5 and Mythos 5, for “foreign national” users. Because nationality checks are difficult at API scale, Anthropic shut the models down globally, the source material says. The directive and its implementation were not publicly detailed in full in the provided material.
At the lunch, Amodei argued for a U.S.-led bloc of democratic countries with managed access to frontier models, limited chip flows to China and joint work on cyber, biosecurity and intelligence risks. Hassabis backed a Western coalition, while Altman called for an international forum on testing standards and public-sector oversight, according to the same material.
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 Control Over Access
Europe’s demand is practical: it wants model access that survives U.S. domestic security decisions. If hospitals, banks, factories, schools or public agencies build workflows around American frontier models, a foreign export order can become a local service outage.
The episode also adds force to Europe’s push for technological sovereignty. Macron and European officials have tied that agenda to large-scale compute, power and chip projects, including a cited €420 billion package, AI gigafactories and European model programs such as CADA. Those plans are political and industrial responses to a dependency that businesses experienced as an operational cutoff, not only a policy risk.
The three CEOs can offer cooperation, standards and commercial access, but the source material’s central tension is that they do not control the U.S. government’s export switch. That gap explains why Europe is asking for enforceable access rights, local compute and a voice in where AI infrastructure is built.
Access Control Systems: Security, Identity Management and Trust Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Shutdown Changed The Talks
The Évian meeting followed a series of AI summits that have moved from safety principles toward economic control and strategic access. The official G7 framing was safe, rapid and effective AI deployment, but the reported Anthropic cutoff gave the session a sharper edge.
European and allied labs were included for a reason. Mistral, Synthesia, Black Forest Labs, Domyn and Sakana AI represented a counterweight to the largest U.S. labs, though none currently has the same global reach as OpenAI, Google DeepMind or Anthropic. Their presence pointed to a European and allied bet on local model capacity, open weights and self-hosted systems.
According to the source material, the summit did not reverse the U.S. restrictions. Instead, leaders discussed a platform for Western democracies within a month, a broader meeting in September, trusted-partner access and common principles on child and youth safety.
The Developer's Playbook for Large Language Model Security: Building Secure AI Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Guarantees Remain Beyond CEO Control
Several points are still not settled. The public record available in the source material does not show the full legal text of the June 12 directive, the exact duration of the Anthropic restrictions, or whether any European customers received exemptions after the shutdown.
It is also unclear what a trusted-partner scheme would cover: API access, model weights, compute reservations, chip sales, security vetting or all of those. Any durable arrangement would require U.S. government approval, not only company promises.
Solutions Architect's Handbook: Kick-start your career with architecture design principles, strategies, and generative AI techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
September Talks Become The Test
The next test is whether the proposed Western democracies platform produces specific rules for access, auditing, cybersecurity cooperation and youth safety before leaders meet again in September. Europe will also need to decide how much of its response can be handled through guarantees from U.S. firms and how much requires European compute and models.
For businesses and public institutions, the near-term issue is contingency planning. The Évian lesson is that frontier AI access can be governed by national security policy as much as by contracts, and that backup models, self-hosted options and jurisdictional risk checks are now part of AI procurement.
trusted AI model deployment platform
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened at the G7 lunch in Évian?
G7 leaders and senior officials met with major AI executives on June 17 to discuss AI deployment, access and safety. The meeting was shaped by reports that a U.S. export-control directive forced Anthropic to shut off its latest models worldwide.
Did the U.S. export order get reversed?
No reversal is confirmed in the source material. The reported ban remained part of the summit fallout, with leaders discussing future access rules rather than an immediate fix.
What does Europe want from Amodei, Hassabis and Altman?
Europe wants reliable access to frontier models, protection against sudden shutdowns, a trusted-partner system, more local compute and model capacity, and stronger child and youth safety rules.
Why can’t the AI CEOs simply promise access?
The companies can offer contracts and cooperation, but export controls are government decisions. If Washington restricts access on national security grounds, company promises may not override that order.
What should companies and public institutions watch now?
They should watch whether September talks produce enforceable access rules, whether exemptions are created for allied users, and whether European model and compute projects move from pledges to operational capacity.
Source: Thorsten Meyer AI
