Anthropic restricts EU access to advanced Claude Mythos cyber model

The restriction leaves European banks and software developers unable to stress-test their defenses against a system capable of identifying high-level vulnerabilities. While OpenAI has proactively shared its ChatGPT 5.5-Cyber model with the EU’s AI Office, Anthropic’s decision to withhold Mythos follows intense pressure from the White House. U.S. officials have expressed concerns regarding the potential weaponization of the model, even considering executive orders to mandate federal vetting of such systems prior to any international release.

European lawmakers are increasingly vocal about the risks this exclusion poses to the region’s critical infrastructure. Paul Timmers, a professor at KU Leuven, warned that selective distribution creates a dangerous bottleneck, effectively tying European security to the strategic decisions of American corporations. Within the European Parliament, members like Stéphanie Yon-Courtin argue that the move undermines claims of strategic autonomy, asserting that Europe cannot rely on Washington to decide which security tools are accessible to its institutions. With European finance sectors still reliant on aging software, the inability to access tools like Mythos heightens the threat of systemic cyber failures across the continent.