The U.S. government has ordered Anthropic to shut off access to two of its most powerful AI models, citing security concerns. The directive forces the company to disable Claude Fable 5 and Mythos 5 for all users worldwide, despite Anthropic’s disagreement with the move.

Mythos, described as an exceptionally capable model for finding software vulnerabilities, was shared only with vetted organizations due to its high-risk potential. Fable 5, released just three days ago, was designed with guardrails but still faced immediate scrutiny.

Anthropic argues that the government’s concerns are overblown and that similar capabilities exist in other models. The company believes the directive is more about a possible jailbreak than genuine security threats. This move comes as Anthropic prepares for an IPO this year, potentially disrupting its business plans.

The irony isn’t lost on observers; while promoting Mythos as uniquely dangerous, Anthropic now faces government scrutiny that could impact its market strategy. OpenAI’s Sam Altman might be enjoying the turn of events, having previously criticized Anthropic's marketing tactics.