European regulators sidelined as Anthropic limits release of powerful new model Mythos

What happened

It has been reported that Anthropic last week restricted access to its latest model, Mythos, handing it only to a small circle of trusted partners while warning the model “outperformed most humans” in finding and exploiting technical vulnerabilities. The move, allegedly intended to give partners time to patch systems, sent a shock through the AI and cyber communities. Big questions followed fast: who gets the keys to frontier AI, and who’s left out in the cold?

Who got access — and who didn’t

It has been reported that Anthropic handpicked a first tier of 12 tech companies, all headquartered in the U.S., including Apple, Microsoft and Amazon, and granted access to about 40 other organizations it did not publicly name. It has also been reported that the company has been in “ongoing discussions with U.S. government officials.” European cyber agencies say they were largely sidelined. Only Germany’s BSI reported dialogue with Anthropic and no full testing; the U.K.’s AI Security Institute, by contrast, did test Mythos and said it “took action on our findings.” The picture is uneven. Messy. And politically awkward.

Why it matters

Experts warned this episode exposes a yawning governance gap. It has been reported that Daniel Privitera of Berlin’s KIRA said Mythos shows how crucial access to frontier capabilities will be — and that Europe currently lacks a plan to secure that access. Yoshua Bengio called it “deeply concerning,” arguing companies shouldn’t be the sole arbiters of public safety. Claudia Plattner of Germany’s BSI warned that whether tools of “such extraordinary power” reach the open market has “profound implications for national and European security and sovereignty.” Emotion? A mixture of frustration and fear — regulators feeling both powerless and outrun as private labs set the pace.

Can Europe patch its way into the conversation before the next model drops? That’s the real test now — policy, trust and technique all on the line.

Sources: politico.eu