NSA Using Anthropic Mythos Despite Pentagon Blacklist
SkimNews Take
Despite official policy, the operational utility of advanced AI models is driving their adoption within defense agencies, creating a de facto integration that outpaces formal risk assessments.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- NSA is using Anthropic's Mythos Preview model — the company's most powerful yet — despite the Department of Defense labeling Anthropic a "supply chain risk," two sources told Axios, with one saying use extends more broadly within the department.
- The Department of Defense moved in February to cut off Anthropic and force its vendors to follow, a case that is still ongoing while the military simultaneously broadens its use of Anthropic's tools.
- Anthropic restricted Mythos access to roughly 40 organizations, arguing the model's offensive cyber capabilities were too dangerous for wider release; the company has publicly named only 12 of those, and the NSA is reportedly among the unnamed recipients.
- Dario Amodei, Anthropic's CEO, met White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent on Friday to discuss Mythos deployment across government; both sides called the meeting productive, with next steps focused on non-Pentagon agencies.
- The Pentagon-Anthropic rift originated in contract renegotiations: DoD demanded Claude be available for "all lawful purposes," while Anthropic insisted on barring mass domestic surveillance and autonomous weapons development — terms some Defense officials cite as evidence the company can't be trusted in wartime.
- The U.K.'s AI Security Institute also has access to Mythos, according to British intelligence counterparts, suggesting allied agencies are already operating under the restricted-access model the U.S. is fighting over.
- Anthropic and the Pentagon both declined to comment; the NSA and the Office of the Director of National Intelligence did not respond to requests for comment.
Why it matters: The contradiction exposes a split-screen federal AI policy: the same department calling Anthropic a national security threat in court is quietly expanding its use of Anthropic's most advanced model, while the White House — by hosting Amodei two days ago — signals it wants the fight settled so other agencies can adopt Mythos without DoD's legal blockade hanging over them.
