NSA taps Anthropic’s restricted Mythos model amid Pentagon‑Anthropic dispute
Key Points
- NSA is using Anthropic's Mythos Preview to scan for cyber vulnerabilities.
- Mythos was withheld from public release due to its offensive capabilities.
- Anthropic limited access to roughly 40 vetted organizations, naming only a dozen publicly.
- The Department of Defense called Anthropic a "supply chain risk" after the firm refused unrestricted Pentagon access.
- Anthropic's refusal to provide Claude for mass surveillance and autonomous weapons sparked the Pentagon dispute.
- UK's AI Security Institute also has access to Mythos.
- CEO Dario Amodei met with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent; meeting described as productive.
- Anthropic declined comment on NSA usage; TechCrunch's request for NSA comment went unanswered.
The National Security Agency has begun using Mythos Preview, Anthropic’s advanced AI model that the company kept out of public release. The move follows a clash with the Department of Defense, which had labeled Anthropic a supply‑chain risk after the firm refused unrestricted Pentagon access to its tools. Anthropic limited Mythos to roughly 40 vetted organizations, naming only a handful publicly; the NSA is now among the undisclosed users, employing the model to scan for exploitable cyber vulnerabilities. The development coincides with a tentative thaw in Anthropic’s ties to the Trump administration, highlighted by a recent meeting with senior White House officials.
The National Security Agency is quietly integrating Anthropic’s Mythos Preview into its cyber‑defense arsenal, according to reporting by Axios. Mythos, announced earlier this month as a frontier model built for cybersecurity tasks, never saw a public rollout. Anthropic cited the model’s potential for offensive cyber operations as the reason for restricting its release, and instead offered access to a select group of about 40 organizations.
Among those privileged recipients, the NSA appears to be a key player. Agency officials are reportedly using Mythos primarily to scan networks and systems for exploitable vulnerabilities, a capability that aligns with the model’s design focus on threat detection and mitigation. The United Kingdom’s AI Security Institute has also confirmed it holds access to the model, underscoring the limited but high‑profile nature of Mythos’s early user base.
The NSA’s adoption arrives on the heels of a public dispute between Anthropic and the Pentagon. The Department of Defense labeled the AI firm a "supply chain risk" after Anthropic balked at granting Pentagon officials unrestricted access to its full‑capability models. The friction grew when Anthropic refused to make its Claude system available for mass domestic surveillance and autonomous weapons development, prompting the DoD to question the firm’s reliability as a defense supplier.
Despite the tension, Anthropic’s relationship with the current administration shows signs of improvement. Last Friday, CEO Dario Amodei met with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent. Sources said the meeting was productive, hinting at a possible easing of the earlier standoff. Anthropic declined to comment on the specifics of the NSA’s usage, and TechCrunch’s request for a comment from the agency went unanswered.
Mythos’s limited distribution raises questions about how the government balances the need for cutting‑edge AI tools with concerns over security and control. By restricting the model to a handful of vetted partners, Anthropic aims to mitigate the risk of its technology being weaponized while still offering a powerful resource to allies deemed trustworthy. The NSA’s involvement suggests the agency values the model’s advanced scanning capabilities enough to navigate the surrounding controversy.
Industry observers note that the episode reflects a broader trend of U.S. intelligence and defense agencies turning to private‑sector AI innovators for specialized tools. As AI models become more capable, the line between defensive and offensive applications blurs, prompting agencies like the NSA to seek solutions that can keep pace with evolving cyber threats. Anthropic’s cautious rollout of Mythos may serve as a template for future collaborations, where access is granted on a need‑to‑know basis rather than through open releases.
While the NSA’s use of Mythos remains largely under the radar, the development signals a shift in how the intelligence community approaches AI‑driven cybersecurity. The model’s ability to quickly identify vulnerabilities could accelerate threat remediation and strengthen national cyber defenses, even as policymakers grapple with the ethical and security implications of deploying such powerful technology.