US government bans Fable 5 over a cyber jailbreak
- https://www.youtube.com/watch?v=W-kM56mf-Nw
- Original title: BREAKING: Fable and Mythos have been taken down for security concerns.
Breaking-news reaction: the US government issued an export-control directive forcing Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national (even foreign-national employees), so Anthropic pulled both models offline for everyone. Other models unaffected.
Citing national-security/export-control authority, the directive landed at 5:21pm ET on a Friday; Anthropic disabled Fable/Mythos within hours because they can't yet verify citizenship. The government's concern: a demonstrated jailbreak (essentially "read this codebase and fix the flaws") that surfaces software vulnerabilities. Anthropic pushes back — the found vulns are minor and other public models (incl. GPT-5.5) find them too without any bypass — and argues recalling a commercial model over a narrow non-universal jailbreak would halt all frontier deployment industry-wide.
Theo connects it to Dario Amodei's "policy on the AI exponential" essay: Mythos-class models are now tools of national strategic consequence (real cyber risk), and policy is reacting fast but bluntly. He speculates it may be retaliatory after Anthropic lobbied against blocking state AI laws. He personally lost money on Fable subs bought for non-US employees, and signs off mourning three days of Fable access while noting OpenAI's next models are likely imminent.