Anthropic released Fable, a publicly available version of its Mythos model, two months after declaring Mythos too dangerous to ship. Days later, the U.S. government issued an export control directive suspending all access to both Fable 5 and Mythos 5 for foreign nationals, citing a jailbreak reportedly flagged by Amazon, an Anthropic investor and inference partner. Anthropic received the directive at 5:21pm ET and disputes its necessity, arguing the demonstrated jailbreak only surfaced minor, previously known vulnerabilities also reachable by other public models. Senior Anthropic staff are currently in Washington seeking resolution.
The model itself matters here. Fable produced a subjective quality jump comparable to GPT-4 and Grok 4, making GPT 5.5 and Opus 4.8 feel dated by comparison. That suggests Fable is downstream of a new pre-training run, not an incremental update. Whether or not this specific jailbreak justifies a federal shutdown is almost beside the point: the author argues the structural conflict between Anthropic and U.S. government authority was already inevitable, and the next model will raise the same fight if this one does not.
The piece uses this episode to examine something more interesting than the legal dispute: why a company that publicly claims to fear its own technology keeps releasing it anyway. The answer involves Anthropic's specific economic position, the billions lost by frontier labs to open-source distillation, and why safety positioning is not just ethics but competitive strategy. The reasoning that connects those pieces is worth reading in full.
[READ ORIGINAL →]