ENISA had less than two weeks with Anthropic’s Mythos model before the U.S. government pulled the plug.
On June 12, a U.S. export control directive suspended all foreign access to Fable 5 and Mythos 5. The stated reason: a third party had reportedly jailbroken Fable 5 by asking it to read a codebase and patch software vulnerabilities. Anthropic called it a narrow, non-universal technique that wouldn’t justify recalling deployed models. The government didn’t agree. Both models went dark globally on the same day.
That left Brussels in an awkward spot. European Commission Executive Vice President Henna Virkkunen flew to Washington this week and raised the ban directly with the Trump administration and with Anthropic. ENISA had been the first EU body admitted to Anthropic’s Mythos testing program, joining a select group of U.S. federal agencies and major companies. Less than two weeks later, it was out.
This is the live drill the EU’s AI sovereignty argument has been waiting for. Virkkunen said Europe needs these models to detect weaknesses in ICT supply chains and can’t rely on single non-European companies for that capability. The export control directive demonstrated exactly what that dependency looks like at 5:21 PM on a Friday. The EC’s action plan on AI and cybersecurity, scheduled for adoption next month, will be written with a documented case study of access revoked overnight.
ECB data released this week showed only 7% of euro area firms use AI intensively. Europe’s capability gap and its access problem just arrived at the same address.
Nathan Zakhary