Anthropic has blocked all customers from accessing Fable 5 and Mythos 5

Anthroic has disabled all of its customers’ access to Fable 5 and Mythos 5 to ensure compliance with an order received from the government on Friday, June 12. All other models and his chatbot Claude are not affected. The company said in its announcement that the US government is asking it to stop all foreign nationals from accessing its recently released AI models, regardless of whether they are in or outside the US and even if they are Anthropic employees, citing national security concerns.

While the US government hasn’t specified these concerns, Anthropic believes it’s because the government has heard about a method to jailbreak Fable 5. The company just released the Fable AI model, which is designed to introduce many of Mythos’ capabilities to the public, on June 9. If you recall, Mythos is her state-of-the-art cybersecurity model that is only available to her Project Glasswing partners. The capabilities of the Fable « exceed » any previous model that Anthropic has released. It beats Pokémon FireRed during the company’s tests, for example, until Claude failed to win Pokemon Redthe original game it’s based on.

Anthropic listed the measures it took to ensure Fable was secure in his post. It said it had put strong safeguards in place to « reduce the likelihood that Fable could be misused for cybersecurity-related tasks » and added that its « safeguards are so strong that many users complain they are overbroad. » The company also explained that no vendor can provide perfect resistance to jailbreak attempts and each model is vulnerable to jailbreaks made specifically for it. « We aimed to make jailbreaks either narrow (in the case of non-universal jailbreaks) or very expensive to produce (in the case of universal jailbreaks) and combine this with comprehensive monitoring to quickly detect and stop any successful attacks, » they said of their defense strategy.

The government apparently provided the company with oral evidence of a potential narrow, non-universal jailbreak that an unnamed individual shared with officials. Anthropic promised to share more details in the next 24 hours, but clarified that it does not agree that a potential jailbreak should be the reason for recalling a commercial model.

« As we have stated publicly, we believe that the government should have the ability to block dangerous implementations as part of a legal process that is transparent, fair, clear, and based on technical facts, » wrote Anthropic, which has been vocal about its warnings about the need for more AI oversight. “This action does not adhere to these principles.