According to a report from Axios, the National Security Agency is reportedly utilizing Mythos Preview, a recently announced model from Anthropic that was not made available for public release. This development follows by just a few weeks the Department of Defense—the NSA's parent agency—designating Anthropic as a "supply chain risk." That label came after the company declined to grant Pentagon officials unrestricted access to the complete capabilities of its AI model.

Earlier this month, Anthropic introduced Mythos as a frontier model specifically engineered for cybersecurity applications. However, the company stated that the model's potential for enabling offensive cyberattacks was too significant to allow a public launch. Consequently, Anthropic restricted access to Mythos to approximately 40 organizations, only about a dozen of which have been publicly identified.

The NSA seems to be among the recipients whose names were not disclosed, and it is reportedly employing Mythos mainly to scan digital environments for security vulnerabilities that could be exploited. The UK's AI Security Institute has also verified that it has been granted access to the model.

This expanded use of Anthropic's tools by the U.S. military is occurring even as the government argues in legal proceedings that such AI tools could pose a threat to national security. The initial disagreement with the Pentagon stemmed from Anthropic's refusal to adapt its Claude model for widespread domestic surveillance programs and for use in autonomous weapons development.

The NSA's reported access to Mythos coincides with an apparent warming in relations between Anthropic and the current presidential administration. Last Friday, Anthropic's CEO, Dario Amodei, held a meeting with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent. The White House described the discussions as productive. Anthropic has declined to provide any comment on these matters.