US government orders Anthropic to disable Claude Fable 5 and Mythos 5 models
Anthropic has complied with the order but argues the action is disproportionate and could stifle innovation across the AI sector.
The US government has issued an immediate directive requiring Anthropic to disable access to its Claude Fable 5 and Claude Mythos 5 models, citing national security concerns. The order, received by the company on Friday at 5:21 pm ET, mandates the shutdown of both systems for all users worldwide, extending beyond the traditional scope of export controls that typically target foreign nationals. Anthropic confirmed its compliance via X but expressed strong disagreement with the rationale behind the move.
The directive appears to be driven by a reported narrow, non-universal jailbreak of Claude Fable 5. According to Anthropic, the government’s concern centres on prompts designed to have the model read specific codebases and identify software flaws. The company noted that the authorities have provided only verbal evidence of this vulnerability, rather than formal documentation. Anthropic argued that the capability to identify such flaws is not unique to its technology and is already present in other publicly accessible models, including OpenAI’s GPT-5.5.
Claude Mythos 5 was previously restricted to a controlled programme known as Project Glasswing, which involved approximately 50 vetted organisations such as Amazon, Apple, Google, Microsoft, and CrowdStrike. The model was kept tightly restricted due to its exceptional ability to find security vulnerabilities in software. Claude Fable 5 was released commercially just three days prior to the shutdown as a version of Mythos fitted with guardrails to block responses in high-risk areas like cybersecurity and biology.
Anthropic maintains that its safeguards rely on independent classifier systems separate from the model itself. The company stated that these systems should prevent dangerous outputs even if a refusal is bypassed. The firm warned that enforcing this standard could effectively halt new model deployments across the AI industry, arguing that recalling a commercial model deployed to hundreds of millions of people based on a narrow potential jailbreak is disproportionate.
The action has drawn attention to the strategic positioning of both Anthropic and its rival, OpenAI. OpenAI CEO Sam Altman previously characterised Anthropic’s restriction of Mythos as fear-based marketing, suggesting that highlighting the model’s danger attracted the scrutiny that has now led to its shutdown. Both companies are widely expected to pursue initial public offerings in the near future, with regulatory outcomes potentially impacting their market trajectories.
