US Orders Anthropic to Pull Fable 5 and Mythos 5 Over Jailbreak Concerns

Anthropic pulled its Fable 5 and Mythos 5 AI models for all customers June 12 after the US government ordered the shutdown on national security grounds, citing a reported jailbreak of the system.

The company said it received the directive at 5:21 p.m. ET. The government’s letter did not explain its specific national security concern, according to Anthropic.

The government believes someone found a way to jailbreak Fable 5, Anthropic said. The company reviewed a demonstration of the technique and said it uncovered only minor, previously known vulnerabilities — the kind other publicly available models can find on their own.

Anthropic said it is complying with the order despite disagreeing with it. The company said no other models are affected.

Before Fable launched, Anthropic ran thousands of hours of adversarial testing alongside the US government, the UK AI Security Institute, and outside organizations. That process showed Fable’s safeguards held up better than any model previously put into production, the company said.

Anthropic said perfect jailbreak resistance is out of reach for any model provider right now. So it designed Fable 5 to limit the damage — making bypasses as narrow and difficult as possible, while running continuous monitoring and holding 30 days of customer data to track and shut down any successful attacks.

The government provided only verbal evidence of a narrow, non-universal jailbreak, Anthropic said — essentially asking the model to read a codebase and correct software flaws. That level of capability is available in other models, including OpenAI’s GPT-5.5, and is used daily by cybersecurity defenders, the company said.

“Would essentially halt all new model deployments for all frontier model providers.” — Anthropic

Anthropic said applying the government’s apparent standard across the industry would essentially halt all new model deployments for all frontier model providers.

The company said it believes the government should be able to block unsafe AI deployments through a process that is “transparent, fair, clear, and grounded in technical facts.” It said this directive does not meet those standards and that it is working to restore access as soon as possible.