At G7, euro AI sovereignty push intensifies after US blocks Anthropic models

But developing models within EU will require hard choices

Europe’s reliance on US AI labs has come under fresh scrutiny after the Trump administration forced Anthropic to disable access to Fable and Mythos, prompting renewed calls for a coordinated European AI strategy.

At the G7 summit in France this week, leaders discussed a proposal to grant select “trusted partners” access to advanced AI systems developed by U.S. firms such as Anthropic. The discussions highlight growing concern among allied nations about restrictions on the use of cutting-edge AI models by non-US countries and businesses.

The talks follow Anthropic’s decision late last week to disable access to its most advanced models, Fable 5 and Mythos 5, for all non-US nationals. The decision came after a directive from Donald Trump ordering the company to block foreign nationals from accessing top-tier AI systems on national security grounds.

Wake-up call for Europe

For European policymakers and business leaders, the abrupt loss of access has reinforced a long-standing concern, that of strategic dependence on foreign AI infrastructure.

Speaking a day later, Maurice Lévy, chairman of Publicis Groupe, warned that European companies are dangerously exposed to external providers that can “flip the switch” on critical technologies without notice. Speaking at VivaTech in Paris, Lévy called for the creation of a €100 billion pan-European AI fund led by France and Germany to strengthen the region’s technological autonomy.

“Artificial intelligence must be treated as a strategic priority,” Lévy said, arguing that dependence on non-European providers could undermine competitiveness, and even threaten the survival of some firms.

Security, competition and control

The dual use capability of the latest generation of models, particularly those from Anthropic, means that they are a geopolitical issue as opposed to a purely economic one.

Models like Mythos, designed to identify vulnerabilities in software, are seen as double-edged tools. Whilst they can strengthen cybersecurity defences, experts warn they could also accelerate sophisticated cyberattacks by identifying vulnerabilities for threat actors to exploit.

The proposed “trusted partners” framework could offer a partial workaround, allowing allied nations to retain access for defensive and research purposes. However, details remain hazy, including whether eligibility would apply to countries, companies, or both.

From access to autonomy

While initiatives to boost tech collaboration and sovereignty across the EU have been discussed for years, recent events appear to have hardened hearts and injected some urgency into those efforts. EU leaders are increasingly framing AI not just as an innovation race, but as a matter of economic resilience and strategic control.

However, the EU must come to terms with the fact that AI sovereignty is about a great deal more than the models themselves. it has less leverage than its size as a trading bloc would suggest. The EU currently hosts around 5% of global datacentre capacity, compared with 80% with the US. If the US wants to develop and host models on its own soil, it's going to have to change these proportions – quickly