Anthropic has restricted public access to its AI models, Fable 5 and Mythos 5, following export restrictions imposed by the Trump administration due to security concerns. The White House's decision was influenced by suspicions that a China-linked group may have accessed the Mythos 5 model. A source familiar with the matter indicated that this development highlights ongoing national security debates regarding Chinese access to American AI technologies.
Concerns regarding a potential method to bypass the models' security features prompted discussions between the administration and Anthropic's CEO, Dario Amodei. Reports indicate that Amodei declined to address the security issues, leading to the export restrictions being enacted. David Sacks, the White House AI czar, stated that the restrictions were implemented reluctantly after Anthropic prioritized the availability of the consumer model over safety.
The restrictions prevent access to the models for customers outside the U.S. and foreign nationals within the country. Anthropic responded by removing access to Fable 5 and Mythos 5 to ensure compliance with the directive from Commerce Secretary Howard Lutnick.
The Mythos model gained attention in April when it was released on a limited basis through Project Glasswing, aimed at securing critical software. However, concerns about its public release were validated when Amazon researchers reportedly jailbroke the model, prompting further discussions at the White House involving senior officials, including President Trump.
Katie Moussouris, CEO of Luta Security, noted that Amazon researchers used prompts to identify vulnerabilities in the AI model. An Amazon spokesperson stated that it is common for governments to seek their counsel on security risks, but details of these discussions are not shared publicly.
Anthropic expressed disagreement with the administration's decision, arguing that the finding of a narrow potential jailbreak should not warrant recalling a commercial model that is widely deployed. The company warned that applying such a standard could halt new model deployments across the industry. Anthropic recently filed to go public, and the two models were released publicly just days prior to the restrictions.