The artificial intelligence company Anthropic communicated this Friday that it is interrupting access to its most advanced systems, "Claude Fable 5" and "Mythos 5", for foreign users, in compliance with an order from the United States Government that limits their use for "national security" reasons.
In an official note, the US firm detailed that the White House directive does not specify the technical reasons for the measure, although they interpret that the authorities believe they have detected a way to bypass the model's security barriers.
"Today we received the government's directive. The letter did not specify the reasons for their national security concerns. We understand that the government believes it has discovered a method to circumvent, or unlock, Fable 5," they reported.
The company has stressed that this model already incorporated strict controls to minimize the risk of misuse in cybersecurity areas, and has reproached the Executive for only having communicated the alleged vulnerabilities verbally, without detailed documentation.
"As we have publicly stated, we believe that the Government should have the power to block insecure deployments, through a transparent, fair, clear, and technically based legal process. This action does not conform to those principles," they denounced.
The Government's intervention comes just 48 hours after Anthropic presented "Claude Fable 5" as the first operational model of the "Mythos" family, a high-performance technology distributed only in a limited way to entities participating in the "Project Glasswing" initiative.
To reduce the risk of malicious uses, Anthropic had incorporated a containment system that automatically restricts responses to sensitive requests related to cybersecurity, biology, or chemistry. In such cases, instead of generating potentially dangerous information, the query is internally redirected to an AI model with lower capabilities.
With this approach, the company intended to offer general assistance in these disciplines, but without providing sensitive data that could be used to launch large-scale cyberattacks or to develop biological weapons.