Anthropic is disabling two of its newly launched AI models, Claude Fable 5 and Mythos 5, following an export control directive from the US government. The order, received Friday afternoon, cites national security concerns and impacts access for all customers to ensure compliance.
This move comes amid ongoing friction between Anthropic and the US government. Earlier this year, the Department of Defense flagged Anthropic as a "supply chain risk," which restricted government agencies and contractors from using its technology. Anthropic responded by filing lawsuits against the administration.
Claude Fable 5, released publicly on Tuesday, is a version of Anthropic's Mythos AI model with built-in safeguards against questions related to cybersecurity, biology, and chemistry. The company stated it collaborated with the US government on this release, following a limited rollout of the Mythos Preview AI model in April aimed at helping organizations bolster their defenses against potential exploitation by bad actors.
In a blog post, Anthropic revealed the US government's letter at 5:21 pm ET, noting it lacked specific details about the national security concern. The company's understanding is that the government believes a method to bypass, or "jailbreak," Fable 5 has been discovered. Anthropic reviewed a demonstration of this technique, identifying minor, previously known vulnerabilities that are also discoverable with other publicly available models.
Anthropic argued in its post that Claude Fable 5 has strong safeguards and that the identified jailbreak is narrow, not making an attacker significantly more dangerous than with other AI models. The company stated the government has only provided verbal evidence of a potential, non-universal jailbreak involving the model reading and fixing code, a technique one potential jailbreaker shared with the government.
Anthropic CEO Dario Amodei has previously expressed support for a structured government process to block unsafe AI models. However, in Friday's blog post, Anthropic contended that this recent government action does not align with those principles.
