The U.S. National Security Agency is reportedly using Mythos Preview, Anthropic’s high-end AI security model, even as the Department of Defense continues to classify Anthropic as a “supply-chain risk.” According to an April 19 Axios report, one part of the U.S. national security apparatus is testing the company’s tools while another is fighting to keep Anthropic models out of government systems.
Mythos access remains tightly restricted
Mythos is not a standard Claude release. Anthropic has limited access through its Project Glasswing group to roughly 40 organizations, including Amazon, Apple, Google, Cisco, CrowdStrike, JPMorgan, Microsoft, and Nvidia. The model is used to scan internal environments for exploitable security flaws. Based on the information cited in the report, Mythos has identified thousands of high-severity vulnerabilities across major operating systems and browsers, which helps explain why Anthropic has kept it on a whitelist instead of making it broadly available.
What remains unclear is how the NSA obtained access, and whether it formally joined Glasswing. Axios said the agency’s specific use of Mythos is not known. That gap matters. It leaves the central question unanswered.
Legal setbacks did not stop talks in Washington
Anthropic sued the Department of Defense in March, challenging the government’s decision to place the company under a supply-chain risk designation. On April 8, a federal appeals court denied Anthropic’s emergency request to pause the restriction, handing the company an early loss in court.
Outside the courtroom, the picture shifted. On April 17, Anthropic CEO Dario Amodei met at the White House with Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. The topic was how Mythos could be used across government agencies. The report also said the White House is pushing to expand federal access to Claude Mythos, while the Defense Department wants Claude to be available for “all lawful purposes.” A separate April 16 report said the White House had already stepped into negotiations over usage terms between Anthropic and government bodies.
Usage limits clash with the NSA’s core mission
Anthropic’s stated restrictions on Mythos are direct: it cannot be used for mass domestic surveillance or for autonomous weapons development. Those limits sit awkwardly beside the NSA’s role in signals intelligence and surveillance. If the agency does have authorized access, the key issue is not simply whether the tool is being used, but what exactly it is being used for.
The dispute now shows three separate lines inside Washington. The Pentagon is keeping Anthropic under a risk label and fighting in court. The White House is trying to broker access terms. The NSA, according to the report, is already testing the tool. Taken together, the episode points to a government still operating without a single settled position on AI capability, security controls, and deployment.