ENISA got Anthropic’s Mythos 5 more than three months late, and still not Mythos 5.1
The European Commission says ENISA, the EU's cybersecurity agency that tests threats and tools for member states, has been granted access to Anthropic's Mythos 5 and is testing it now, Luca Bertuzzi at Euronews writes. Access arrived more than three months after Mythos 5 went to selected partners.
Mythos is Anthropic's restricted high-capability cyber model line. Mythos 5 was previewed in April and can identify and exploit cyber vulnerabilities. Access runs through Project Glasswing, Anthropic's invite-only defensive-cyber partner program, which Euronews puts at about 50 organizations in April, then roughly 200 partners after June expansions.
Commission tech-sovereignty spokesperson Thomas Regnier called the engagement with Anthropic constructive. He also said ENISA previously received OpenAI's GPT-5.6-Cyber and has been granted access to OpenAI's GPT-6 Astra.
A Bloomberg rewrite via CNBC-TV18 adds that Anthropic proposed ENISA access in late May, then months of wrangling followed over type and scope. Anthropic declined to comment.
ENISA will not have access to Mythos 5.1, the newer iteration with more permissive safeguards, per the Commission spokesperson. Britain's AI Security Institute, which had been among the first non-US entities to stress-test the original Mythos, also was not given 5.1 access, according to a person familiar with the matter via the Bloomberg rewrite.
US export controls later barred non-American users worldwide, including Anthropic employees outside the United States, Euronews notes. Commission concern about a US "kill switch" is that framing of export-control politics, not a documented product feature name.
Getting the older restricted model after a public fight looks like progress from the Commission's chair. Sitting at the current table would mean Mythos 5.1.
Europe got yesterday's keys. The newer set stayed on the US side of the Glasswing door.
—