Anthropic blocks access to Fable 5 and Mythos 5 following government directive
US tech firm Anthropic has disabled customer access to its two latest AI models after the government ordered it to suspend foreign nationals' access, citing national security concerns.
Anthropic, an artificial intelligence company, announced that it has cut off all customer access to its Fable 5 and Mythos 5 models. The move is intended to comply with a U.S. government directive issued on Friday, June 12, based on national security concerns.
The order requires Anthropic to suspend access for foreign nationals, including its own employees, to these newly released AI models, regardless of whether they are inside or outside the United States. Other Anthropic models and its Claude chatbot remain unaffected.
While the government did not specify the exact concerns, Anthropic believes the reason is a discovered method to jailbreak Fable 5. Fable 5 was launched on June 9 and designed to bring many of Mythos 5's capabilities to the public. Mythos 5 is a state-of-the-art cybersecurity model available only to Project Glasswing partners. Fable 5's capabilities exceed any previous Anthropic model—for instance, it beat Pokémon FireRed during testing, whereas Claude failed to beat the original Pokémon Red.
Anthropic stated that it had implemented strong safeguards to reduce the likelihood of Fable 5 being misused for cybersecurity-related tasks. The company noted that many users have complained these safeguards are overly broad. However, Anthropic acknowledges that no provider can guarantee perfect resistance to jailbreak attempts, and every model is vulnerable to specially crafted attacks. Its defense strategy aimed to make jailbreaks either narrow (non-universal) or very expensive to produce (universal), combined with thorough monitoring to quickly detect and shut down successful attacks.
The government reportedly provided verbal evidence of one potential narrow, non-universal jailbreak shared by an unnamed entity. Anthropic promised to share more details within the next 24 hours but stressed that it disagrees that a potential jailbreak warrants recalling a commercial model. The company, which has been vocal about the need for AI safety, wrote that it believes the government should have the ability to block unsafe deployments as part of a transparent, fair, and technically grounded statutory process. It added that this action does not adhere to those principles.
