US-EU AI cooperation in focus at G7 Summit

Ursula von der Leyen urged the US and Europe to deepen AI cooperation at the G7 Summit, framing it as a shared economic and safety imperative. She noted the two regions together represent about 70% of the world AI market and called for joint acceleration of adoption while introducing powerful models safely. The push comes after a US government order directed Anthropic to restrict its most advanced models for foreign nationals, leading Anthropic to shut down Fable 5 and Mythos 5 entirely rather than verify nationality per user.

What happened

Speaking after a roundtable at the G7 Summit in France, von der Leyen described AI as "the most important technology of our time" and stressed that the US and EU have a shared interest in both adoption and safety. She argued that because the two economies use each other's trusted technology and have interconnected financial systems, AI safety is a mutual interest.

Earlier that week, the Trump administration sent a letter directing Anthropic to suspend access to its Fable 5 and Mythos 5 models for foreign nationals, citing national security concerns. Anthropic responded by shutting down both models for all customers, arguing that verifying every user's nationality was impractical. French Prime Minister Sebastien Lecornu seized on the incident, arguing Europe "cannot rely on the goodwill of certain partners" who can unilaterally cut off access.

Von der Leyen responded by calling for G7 countries to exchange information and cooperate via independent entities that evaluate models before market release. She also emphasized that Europe wants "our own AI future, not in isolation ... but together with our trusted partners".

Why AI builders should care

For teams shipping AI products, the G7 debate signals that cross-border model access is becoming a regulatory and geopolitical variable. The Anthropic shutdown shows that even a well-funded frontier lab can pull models globally overnight when faced with compliance obligations that are impractical to meet. If you depend on US-hosted frontier models for your EU customers, you now have a concrete example of supply chain fragility.

At the same time, von der Leyen's call for independent model evaluations before market release points toward a future where regulatory gatekeeping could delay or block model availability in Europe. Builders should monitor whether such evaluations become mandatory under the EU AI Act or bilateral US-EU agreements, as that would affect release timelines and testing costs.

The framing of the US and EU as "at par in terms of AI adoption" suggests that Europe does not see itself as lagging behind in usage, even if it lacks its own frontier model ecosystem. For indie developers and product teams, this could mean a more level playing field for distribution, but also more regulatory friction for any product relying on US API providers.

Practical implications

• Diversify model access. The Anthropic case is a reminder to avoid single-vendor lock-in for critical inference paths. Consider having fallback models from other labs or locally hosted alternatives for EU deployments.
• Audit compliance readiness. If your product uses Anthropic's latest models, note that access can be revoked without warning. Plan for model replacement scenarios and test on other providers.
• Watch for evaluation requirements. Von der Leyen specifically mentioned independent entities evaluating models before market release. This could morph into a certification regime similar to CE marking. Builders should preemptively document model testing and safety procedures.
• Expect sovereignty investments. French PM Lecornu's comments and Europe's pursuit of its own AI future suggest government backing for European AI infrastructure. Keep an eye on compute grants, cloud credits, or API subsidies for EU-hosted models.

Caveats

Von der Leyen's statements are policy aspirations, not binding agreements. No concrete timeline or mechanism for the proposed independent evaluations has been disclosed. The Anthropic incident, while real, is a single event and may not represent a broader trend in US export controls. Additionally, the G7 discussions also covered other contentious topics like Ukraine and Iran, and AI cooperation may be traded off against other geopolitical priorities. Builders should treat these signals as directional but not actionable until formal agreements or regulations emerge.