Anthropic has suspended access to its newly launched Fable 5 and Mythos 5 artificial intelligence models after receiving a U.S. government export control directive tied to national security concerns.
Summary
- Anthropic suspended Fable 5 and Mythos 5 after receiving a U.S. export control directive.
- The company said officials cited national security concerns linked to a potential jailbreak method.
- The move comes days after the launch of the new AI models and amid a major infrastructure expansion push.
According to a statement published by Anthropic on Friday, the company received the directive at 5:21 p.m. ET, instructing it to block access to Fable 5 and Mythos 5 for all foreign nationals, regardless of whether they are located inside or outside the United States. The order also applied to foreign-national employees working at Anthropic.
Faced with the directive, Anthropic said it disabled both models for all users to ensure compliance with the government’s requirements. The company added that its other models, including Opus 4.8, remain available and are not affected by the restrictions.
“We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users.”
The move comes only days after Anthropic introduced Fable 5 as a generally available Mythos-class model and released Mythos 5 for a limited group of approved cybersecurity and infrastructure users.
According to Anthropic, Fable 5 was designed to handle longer and more complex tasks than previous Claude models and delivered strong performance across software engineering, scientific research, finance, vision, memory, and knowledge work.
Government concerns center on potential model jailbreak
While authorities did not provide detailed evidence supporting the order, Anthropic said it believes the government is concerned about a possible jailbreak technique that could bypass some of Fable 5’s safeguards.
According to Anthropic, officials have so far presented only verbal evidence of what the company described as a narrow, non-universal jailbreak. The company said the reported method involves asking the model to analyze a specific codebase and identify or repair software vulnerabilities.
Anthropic explained that a non-universal jailbreak differs significantly from a universal jailbreak because it does not broadly remove a model’s safety protections across a wide range of tasks.
“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. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
At the same time, Anthropic said it is working with authorities and believes the directive may have resulted from a misunderstanding. The company stated that it is seeking to restore access as quickly as possible.
Infrastructure expansion continues despite model restrictions
Even as access to Fable 5 and Mythos 5 remains suspended, Anthropic continues to expand its computing capacity for future AI systems.
As reported by crypto.news, private credit firms Blackstone and Apollo Global Management are syndicating approximately $36 billion in financing to support Anthropic’s next phase of infrastructure spending. Reuters reported that the funds will be used to acquire custom tensor processing unit chips from Google, backed by Broadcom technology, which Anthropic plans to lease for its AI operations.
Separately, Anthropic has been urging governments to establish rules for frontier AI systems as model capabilities advance. The company has proposed policy measures covering dangerous deployments, independent evaluations, cybersecurity safeguards, and economic preparation for workers affected by AI adoption.
Those policy recommendations now arrive as Anthropic finds itself at the center of one of the most significant government interventions involving a newly released frontier AI model.
Credit: Source link
