U.S. Government Orders Anthropic to Suspend Two Leading AI Models Over Security Concerns
Anthropic said it has disabled its Claude Fable 5 and Claude Mythos 5 models worldwide after a U.S. export-control directive cited national-security risks, while the company disputes the justification.
The United States government instructed Anthropic on Friday to block access to its two most advanced AI systems, Claude Fable 5 and Claude Mythos 5, citing concerns that the models could be exploited for national-security threats. Anthropic confirmed on its X account that it has complied, but the firm argues the action is unwarranted.
The order, received at 5:21 p.m. ET, requires the company to disable the two models for all users globally, not only for foreign nationals as the export-control language suggests. Other Anthropic models remain available.
Claude Mythos 5, described by the company as its most capable model, was initially released to a limited group of about 50 vetted organizations under a program called Project Glasswing. Anthropic said the model can identify software vulnerabilities and was intended for defensive cybersecurity work.
Claude Fable 5, launched three days earlier, is a version of Mythos 5 with additional guardrails intended to block high-risk topics such as cybersecurity and biology, making it suitable for broader commercial use. Independent benchmarks listed it as the most capable publicly accessible AI model at the time of release.
Anthropic’s blog post indicates the government’s concern centers on a reported “narrow, non-universal jailbreak” that allegedly allows the model to read a codebase and point out software flaws. The company says the government has provided only verbal evidence of this issue and notes that similar capabilities exist in other publicly available models, including OpenAI’s GPT-5.5.
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company wrote.
Anthropic added that its safety mechanisms rely on independent classifier systems that operate separately from the model, so even if a user bypasses a refusal, the underlying protections against dangerous outputs remain active.
The directive arrives as Anthropic prepares for a potential initial public offering later this year and has positioned itself as a safety-focused alternative to rival AI developers.
