"They screwed us": Personality clashes sent Anthropic's models offline
TL;DR
Axios describes an escalation between Anthropic and the Trump administration: Mythos and Fable went offline after new export controls, with officials accusing Anthropic of failing to take a cyber executive order seriously. On June 11, Amazon CEO Andy Jassy reportedly warned Treasury Secretary Scott Bessent that Mythos and Fable could be jailbroken. Anthropic says Fable launched with explicit government approval.
Nauti's Take
This looks less like a clean security decision and more like a power and communications crash with real product fallout. Anthropic cannot rely on a safety reputation if officials feel dismissed; in frontier AI, trust is operational infrastructure.
At the same time, the reporting is heavily shaped by administration sources, so treat the blame narrative as contested framing, not a settled postmortem.
Briefingshow
The case shows how quickly AI governance can move from technical risk to trust, status and political communication. Perfect jailbreak resistance is unrealistic, so labs are being judged not only on model hardening but on whether officials believe their risk escalation. Export controls are a fast lever, but a blunt one that can remove major products overnight.
Sources