Brussels Assesses Impact of Mythos and Fable Ban and Urges Washington Not to Discriminate

The European Commission confirmed on Sunday (14) that it is evaluating the practical consequences for European companies and researchers of the U.S. Department of Commerce's order that globally took down Claude Mythos 5 and Claude Fable 5 from Anthropic. In a statement to Reuters, Commission spokesperson Thomas Regnier stated that "contingency measures should not be discriminatory against partners" and called for Europe to strengthen its technological sovereignty.

The order from Secretary of Commerce Howard Lutnick was sent to Anthropic on Thursday (11) and subjected the two models to export controls. As Anthropic is unable to isolate its users by nationality in real time, the company globally shut down Mythos 5 and Fable 5. German universities, French banks, and enterprise software vendors across the continent lost access to the company's more capable models overnight, without prior notice.

What Brussels Can (and Can't) Do

The Commission does not have a direct instrument to reverse an American export control. What is at stake is how Brussels will use this incident to justify ongoing movements. The EU AI Act, in the final stages of implementation, mandates continuity plans for high-risk AI systems, and the shutdown serves as a concrete example of the risk of a single provider. The NIS 2 Directive, in effect since October 2024, requires banks and operators of essential services to document critical supplier dependencies.

Regnier did not mention retaliation, but the chosen framework (technological sovereignty) is the same used to justify the EuroStack program, which received €1.2 billion from the European budget in 2026 for cloud infrastructure and continental AI models. For Mistral, a French provider, and Aleph Alpha in Germany, Anthropic's shutdown provides a ready commercial argument.

Who Was Affected Practically

Capgemini, Sopra Steria, and Atos have been delivering standardized outputs to European financial sector clients on Claude since 2025. Capgemini disclosed in May that about 18% of its generative AI contracts run on some model from Anthropic. Deutsche Bank, which confirmed in February a partnership with Anthropic for research automation, lost access to Mythos 5, a model that had been tested for internal offensive security. UBS and BNP Paribas, with similar pipelines, activated contingencies in Gemini and their own models over the weekend.

In Japan, MUFG and Mizuho, which announced multi-year contracts with Anthropic via AWS earlier this year, froze two pilot projects. In India, delivery hubs from Infosys and TCS serving European clients in operational security reported interruptions in flows dependent on Mythos. Salesforce, the third-largest buyer of tokens from Anthropic in 2026 according to Marc Benioff, is excluded from the newer models of the provider.

The Blind Spot of the AI Act

There is an irony that does not escape regulators in Brussels: the EU AI Act was designed to contain risks from Europe, not to protect European companies from American decisions. Andrea Renda from the Centre for European Policy Studies has been arguing for months that the architecture of the Act has a jurisdictional blind spot. According to Renda, the Act regulates how models can be used in Europe, but it does not guarantee that they will remain accessible in Europe. Anthropic's shutdown transforms this criticism from theoretical to operational.

Some defend the American position. The R Street Institute published an analysis on Saturday classifying the action as "a poorly applied bad idea," acknowledging, however, that the White House operates within its legal authority. The Cato Institute argues that the export control regime has always had broad discretion and that the problem is institutional, not constitutional.

The Message for Suppliers and Buyers

The Commission will not overturn the U.S. decision, but it will use it. Regnier indicated that the next round of AI regulation, expected in the second half of the year, should include more explicit requirements for model portability and continuity clauses in contracts with foreign suppliers. For CIOs in Frankfurt, Paris, and Madrid, this means that the next generative AI tender will require proof of a backup plan before performance proof. For CIOs in Tokyo and São Paulo, the reading is similar: vendor lock-in with a supplier from another jurisdiction has moved from the acceptable risk category to the material risk category.