US government orders Anthropic to pull Claude Fable 5 and Claude Mythos 5, issuing an emergency export control directive that requires the company to suspend access to those models for all foreign nationals. The directive bars any foreign national from accessing the models, whether inside or outside the United States, and includes Anthropic employees among the affected foreign nationals. Anthropic complied by disabling both Claude Fable 5 and Claude Mythos 5 for its entire customer base to ensure compliance, and the company said it is working to restore access soon.
Claude Mythos 5 was released only to select partners and was reported to include fewer guardrails than Anthropic’s publicly distributed offerings, while Claude Fable 5 was publicly available and was specifically cited in reporting as the model referenced in connection with a potential vulnerability. The U.S. government raised concerns about that vulnerability, framing it as a case of “jailbreaking” the models and reportedly providing only verbal evidence about the incident.
The verbal evidence was described as showing a narrow, non-universal jailbreak that essentially consisted of asking the model to “read a specific codebase and fix any software flaws.” Coverage contrasted Mythos 5’s reduced guardrails and partner-only access with Fable 5’s public distribution when describing the reported issue. Officials characterized the reported issue as limited in scope rather than a universal bypass of model protections.
Anthropic complied with the emergency export control directive by disabling Claude Fable 5 and Claude Mythos 5 for its entire customer base while disputing the severity of the government’s finding and calling the directive an overreach. The company said the reported vulnerabilities appear relatively simple and argued that comparable capabilities are already available from other publicly accessible models, explicitly citing OpenAI’s GPT-5.5 as an example.
Anthropic wrote in public statements that “if this standard was applied across the industry,” it “would essentially halt all new model deployments for all frontier model providers.”
The company also said the issue was already widespread across the industry and that access to all other Anthropic models would not be affected; Anthropic added that it is working to restore access to unaffected models as soon as possible.
David Sacks claimed that a highly credible, trusted partner who was testing Claude Fable came forward with a jailbreak of the model’s guardrails. Sacks said that, following that report, an administrator asked Anthropic CEO Dario Amodei to fix the jailbreak or to withdraw the model from deployment, and that Amodei refused that request. Sacks further asserted that Anthropic’s refusal to de-deploy or promptly fix the reported jailbreak contradicted the company’s public statements on AI safety and regulation, referencing a blog post by Amodei. He characterized the internal exchange as evidence of testing the jailbreak and as illustrative of a perceived gap between Anthropic’s public safety commitments and its internal choices.
The U.S. government issued an emergency export control directive ordering Anthropic to suspend access to Claude Fable 5 and Claude Mythos 5 for all foreign nationals, and Anthropic disabled those models for its customer base. Anthropic described the government’s action as an overreach and said the reported vulnerability was industry-wide, and the company said it is working to restore access to other, unaffected models.
