Anthropic has restricted access to its AI models, Fable 5 and Mythos 5, for all customers following a government order issued on June 12. This action affects foreign nationals, including Anthropic employees, irrespective of their location, due to cited national security concerns. The government did not disclose specific details regarding these concerns.
On June 9, Anthropic launched Fable, which builds upon the capabilities of Mythos, its advanced cybersecurity model available only to select partners. The Fable model surpassed previous benchmarks, outperforming Pokémon FireRed during internal tests. In contrast, the Claude chatbot did not achieve similar success against the original Pokémon Red.
To enhance Fable's security, Anthropic implemented robust safeguards aimed at minimizing misuse related to cybersecurity. Despite these measures, the company acknowledged that no model can be entirely resistant to jailbreak attempts. They mentioned that the government provided verbal evidence of a potential jailbreak method from an undisclosed source.
Anthropic plans to release further information within the next 24 hours while maintaining that a potential jailbreak should not warrant the recall of a commercial product.