EU warns Washington over Anthropic AI lockout
Washington abruptly cut European access to Anthropic's top AI models. Brussels says it is assessing the impact and warns against unequal treatment.
In a matter of hours on Friday evening, the Trump administration cut off access for all non-U.S. users — including in Europe — to the two most advanced artificial intelligence models built by Anthropic, the U.S. AI company behind the Claude chatbot: Fable 5 and Mythos 5. Officially justified on national security grounds, the move drew a measured response from the European Commission, the EU’s executive arm, which said two days later it was assessing the decision’s impact on European users — while cautioning that such precautionary steps should not end up treating allies unequally.
This image is used for illustrative purposes only.
At a Glance
Washington ordered Anthropic on Friday to block Fable 5 and Mythos 5 for all non-U.S. users, citing national security.
The European Commission says it is assessing the move’s impact on European users, while cautioning that it should not result in unequal treatment between partners.
The episode revives the debate over Europe’s reliance on top-tier U.S. AI models, days before a G7 summit where Anthropic will be represented.
How the access cutoff unfolded
According to Anthropic’s own account, the company received a letter from the U.S. government at 5:21 p.m. Eastern Time on Friday, ordering it to immediately suspend access to Fable 5 and Mythos 5 for all foreign nationals — whether located inside or outside the United States — as well as for its own foreign employees. Anthropic says it complied within the hour, disabling both models for all non-U.S. customers, while stressing that access to its other products remains unaffected.
U.S. authorities cited the risk of a jailbreak of Fable 5 — a technique used to exploit a model’s weaknesses and push it to generate content its safeguards are designed to block. Anthropic disputes whether the measure is proportionate. The company says the flaw identified is relatively simple and that publicly available models are already capable of detecting it. It argues that such a limited jailbreak risk does not justify pulling a commercial model used by hundreds of millions of people, and notes it has built in safeguards that sharply reduce the risk of the model being misused for cybersecurity purposes. Anthropic apologized to its users and says it is working to restore access.
Brussels takes a measured tone — and starts assessing
On Sunday, Thomas Regnier, the European Commission spokesperson for technological sovereignty, said the EU is witnessing the emergence of a new generation of highly capable AI models that bring major benefits — particularly for cyberdefense — but also raise serious cybersecurity concerns. He said the Commission is taking note of the U.S. decision and assessing its implications, stressing that this is a global issue that goes beyond borders and individual companies, and that precautionary measures of this kind should not result in unequal treatment between partners.
According to Brussels, the episode is further evidence that Europe needs to strengthen its own technological sovereignty — including by leaning on its regulatory framework for artificial intelligence and cybersecurity to manage risks of this kind.
Another chapter in a tense relationship
This is not an isolated dispute. Back in February, President Trump ordered U.S. federal agencies to immediately stop using Anthropic’s technology, after the company and its chief executive, Dario Amodei, pushed back against certain military uses of its models. On his Truth Social platform, Trump said at the time that the government no longer needed Anthropic’s services and would cease doing business with the company altogether, while allowing a six-month transition period for the agencies involved. Anthropic later said it would pursue legal action against the government after being designated a “supply chain risk.”
Friday’s move also drew political reactions in Europe. Jordan Bardella, president of France’s far-right Rassemblement National (RN) party, said the episode was a reminder that artificial intelligence is now a major national sovereignty issue. He called on France to step up support for Mistral AI, the French AI developer often described as Europe’s leading challenger to U.S. labs, as well as for the broader domestic AI sector, arguing that countries that fail to build their own models will remain dependent on foreign powers.
In the UK, Tom Tugendhat, a British member of parliament and former security minister, said the decision illustrated technology’s strategic centrality: disabling models for foreign users, he argued, is not a minor technical hiccup but a sign that sovereignty is now defined as much by code as by weapons.
The real question: can sovereignty exist without independence
The episode fits into a broader pattern of transatlantic tech friction — from U.S. export controls on advanced semiconductors to European debates over U.S. authorities’ access to data stored in the cloud. But it highlights a particular imbalance. The European Union has, in its Artificial Intelligence Act (AI Act), what is widely regarded as one of the most comprehensive regulatory frameworks in the world. Yet that framework gives Brussels no equivalent lever over whether these technologies remain available to Europeans at all. The decision to cut off Fable 5 and Mythos 5 was made unilaterally in Washington, for reasons of U.S. domestic security, with apparently no consideration for how the two models were being used in Europe.
That dependence is not just symbolic. European research labs or cybersecurity firms may well have been among the users of these two models for their advanced vulnerability-detection capabilities — and for them, access disappeared overnight, with no guaranteed timeline for restoration.
That may be the real question this episode raises. The Commission can raise concerns about unequal treatment and point to its regulatory arsenal, but it cannot, as things stand, stop a foreign government from deciding overnight which digital tools Europeans can or cannot use. Bardella’s call to invest more in Mistral AI, however political its motives, points to a concern that some officials in Brussels, in more measured language, might well share:
As long as the world’s most advanced AI models remain American, European digital sovereignty will, by construction, remain a conditional one.
The Bottom Line
On Tuesday, in Évian, France, Dario Amodei is set to join a working lunch with G7 leaders and several major AI players — at the very moment his company is enforcing a directive that locks citizens of several of those countries out of its most advanced models. Whether Brussels turns its concern into a concrete demand for reciprocity, or whether this episode becomes just another wake-up call about Europe’s tech dependence — heard, discussed, and quickly set aside — remains to be seen.
Sources: Euronews