According to Axios, the National Security Agency (NSA) is reportedly utilizing Anthropic’s Mythos Preview model, a system intentionally kept from public distribution. This development follows the Department of Defense (DoD), the NSA’s overarching organization, designating Anthropic as a “supply chain risk” because the company denied Pentagon officials full, unimpeded access to its model’s complete functionalities.
Earlier this month, Anthropic introduced Mythos as a cutting-edge model specifically developed for cybersecurity applications. However, the company stated that the model’s potential for initiating offensive cyberattacks was too high for a public release. Consequently, the AI firm restricted access to Mythos to approximately 40 entities, with public disclosure of only about twelve of these.
The NSA is believed to be one of the unnamed organizations granted access, reportedly employing Mythos mainly for identifying exploitable weaknesses within various systems. Additionally, the UK’s AI Security Institute has verified its own access to the Mythos model.
This increasing adoption of Anthropic’s technologies by the U.S. military occurs concurrently with its legal assertion in court that these very tools pose a threat to national security. The conflict with the Pentagon arose after Anthropic declined to permit the use of its Claude model for widespread domestic surveillance and the creation of autonomous weaponry.
The NSA gaining access to Mythos coincides with a perceived improvement in Anthropic’s relations with the Trump administration. Notably, last Friday, Dario Amodei, Anthropic’s chief executive, held a meeting with Susie Wiles, the White House chief of staff, and Scott Bessent, the Secretary of the Treasury.
Reports indicate the White House characterized this meeting as productive.
TechCrunch sought comments from the NSA, but Anthropic chose not to provide a statement.