NSA spies are reportedly using Anthropic’s Mythos, despite Pentagon feud

The National Security Agency (NSA) is reportedly using Anthropic's restricted cybersecurity model, Mythos Preview, according to an Axios report. This development is notable because it comes just weeks after the NSA's parent agency, the Department of Defense (DOD), officially labeled Anthropic a 'supply-chain risk.' That designation stemmed from Anthropic's refusal to grant Pentagon officials unrestricted access to its AI models' full capabilities for purposes like mass domestic surveillance and autonomous weapons development.

Anthropic announced Mythos earlier this month as a frontier model specifically engineered for cybersecurity tasks. The company stated the model was too capable of launching offensive cyberattacks to be released publicly, limiting access to approximately 40 vetted organizations. While Anthropic has publicly named only about a dozen of these, the NSA and the U.K.'s AI Security Institute are confirmed recipients. The NSA is said to be deploying Mythos primarily for defensive purposes: scanning digital environments to identify exploitable vulnerabilities before adversaries can.

The situation highlights a complex and seemingly contradictory relationship between the U.S. government and Anthropic. While one arm of the national security apparatus (the DOD) is arguing in court that Anthropic's tools threaten national security, another (the NSA) is actively deploying them. This access coincides with a potential thaw in relations with the current administration, following a recent meeting between Anthropic CEO Dario Amodei and senior White House officials.