Claude Mythos: Washington Restricts Access, Europe Responds

Claude Mythos at the Heart of European Discussions

On May 4, the finance ministers of the eurozone dedicated part of their meeting in Brussels to an unusual topic: the Claude Mythos artificial intelligence model developed by Anthropic. Typically, these meetings focus on issues such as operational resilience and banking union, but this time, Mythos took center stage. Bloomberg confirmed that the ministers discussed the case of this model and the lack of European access.

Project Glasswing and the European Absence

Project Glasswing, unveiled on April 7 by Anthropic, brings together eleven partners around Mythos Preview, including giants like AWS, Apple, Google, Microsoft, NVIDIA, CrowdStrike, Palo Alto Networks, Cisco, Broadcom, the Linux Foundation, and JPMorgan Chase. However, no European bank is among these partners, which has sparked reactions in Europe. The Bundesbank opened the discussion at the end of April by publicly calling for European access to the model. The argument is straightforward: a model capable of finding zero-day vulnerabilities in all major operating systems gives structural advantages to attackers who access it, while private defenders of the same tool start at a disadvantage.

The European Central Bank (ECB) followed suit by convening the risk directors of eurozone banks. Christine Lagarde described Anthropic as a "responsible" operator, while warning that the model, in the wrong hands, "could be really problematic."

Diverging Views on Access to Mythos

Switzerland, through its financial regulator FINMA, has expressed reservations about immediate and broad access to Mythos, citing systemic risks. Not because the model would be malicious, but because an offensive capability of this caliber, deployed without appropriate defensive infrastructure, could overwhelm incident response teams. The IMF, via Kristalina Georgieva, added that the international monetary system does not yet have the necessary protections against an AI-driven cyber incident. Thus, the debate surrounding Mythos quickly transformed into a monetary policy issue.

American Asymmetry

Washington's position rests on two main arguments: a model capable of writing exploits cannot be widely distributed without the risk of misuse, and the necessary infrastructure (secure facilities, access controls, oversight) does not yet exist for the broader circle that Anthropic proposes. The White House has therefore blocked the extension to about 70 additional organizations. However, the NSA is already using Mythos under agreements made prior to the controversy, and the U.S. Treasury has also requested access to audit its own systems. This restrictive posture towards the outside and permissive internally, initiated by Trump on April 17 after a meeting between Dario Amodei and the White House, fuels the European reaction.

Future Prospects for Europe

Anthropic has privately indicated that European access would be provided "soon," but no formal agreement has been signed. The Eurogroup did not produce a timeline during its meeting. Three scenarios are circulating: a discreet bilateral arrangement between Washington and Brussels, a months-long stalemate, or a European investment in a comparable sovereign model. The SPRIND program, with 125 million euros to build the first frontier AI laboratories in Europe, is already pointing in this direction. For French and European banks, the equation remains the same: attackers will eventually gain access to this type of capability, through one channel or another, and defenders deprived of equivalent tools will remain structurally behind.