Anthropic forced to disable Fable 5 and Mythos 5 globally after US government jailbreak order

Anthropic was forced to instantly disable its flagship models Fable 5 and Mythos 5 for all global customers following an emergency export control directive from the US government. The trigger was a jailbreak that allowed the models to fix security vulnerabilities in a specific codebase when prompted in a particular way. The government deemed this capability a national security risk under export controls, leading to a complete API shutdown without any transparent appeals process. Anthropic has publicly pushed back, arguing the standard is so broad it would halt all new frontier models if applied consistently, and that the goal ostensibly is to prevent adversaries from using the models for offensive cyber operations.

This incident highlights a critical weakness of relying on centralized API-based AI systems: a single government decree can nuke a model globally over what many consider a trivial prompt. The fact that a model could be banned for helping fix software flaws — a behavior typically encouraged — has shocked many in the AI community. Anthropic admits that if this precedent stands, it would impossible to release any capable model without risking sudden global revocation. The company is advocating for a clearer, more proportional regulatory framework, but for now hundreds of millions of users have lost access. This is exactly why advocates push for local, open-weight models that cannot be remotely disabled.