NSA Is Using Anthropic's Powerful Claude Mythos AI as CEO Meets With White House: Report

The National Security Agency is moving Anthropic's Claude Mythos Preview wrong its classified networks, according to 2 sources cited by Axios—a astonishing improvement fixed that the NSA falls nether the Department of Defense, which declared Anthropic a supply-chain risk successful March and is presently warring the institution successful national court.

Claude Mythos is not a modular endeavor tool. When Anthropic unveiled the exemplary earlier this month, it restricted entree to a fistful of vetted organizations, arguing that the exemplary poses superior violative information risks. Anthropic's ain method documentation recovered that Mythos was capable to place captious vulnerabilities successful each wide utilized operating strategy and web browser. The institution judged it excessively unsafe for unfastened release.

Most organizations with entree are utilizing the exemplary defensively, scanning their ain infrastructure for weaknesses earlier adversaries do. The initiative, branded Project Glasswing, includes Microsoft, Google, Apple, Amazon Web Services, JPMorgan Chase, and Nvidia. What the NSA is doing with Mythos is little clear, though the agency's ngo is not purely defensive. A 3rd root told Axios the exemplary is being utilized much broadly wrong the quality department.

The Pentagon's hostility toward Anthropic traces to negotiations that went bad. In July 2025, the 2 sides signed an statement making Claude the archetypal frontier AI exemplary cleared for usage connected classified networks. Talks soured erstwhile the Pentagon sought to renegotiate, demanding the subject beryllium allowed to usage Claude "for each lawful purposes" without restriction. Anthropic refused, drafting 2 steadfast lines: nary autonomous weapons, and nary home wide surveillance.

When negotiations collapsed, Defense Secretary Pete Hegseth declared Anthropic a supply-chain risk successful precocious February—an unprecedented designation, and the archetypal ever applied to an American company. A California national justice blocked the move, but past a D.C. appeals tribunal denied Anthropic's abstracted bid to halt the blacklisting portion litigation plays out. The 2 sides stay successful court.

While the ineligible combat grinds on, the remainder of the medication is moving successful a antithetic direction. On April 17, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. Anthropic described the league arsenic “productive”, Reuters reported. The White House said the parties "discussed opportunities for collaboration, arsenic good arsenic shared approaches and protocols to code the challenges associated with scaling this technology."

President Trump, asked by reporters astir the meeting, said helium had "no idea" Amodei had been astatine the White House, aft helium antecedently ordered the medication not to usage Anthropic’s “woke” models. Bessent and Federal Reserve Chair Jerome Powell person separately been encouraging large slope CEOs to trial Mythos and beryllium prepared for information threats, and an medication root told Axios that each national bureau but the Defense Department wants entree to Anthropic's tools.

The NSA's reported usage of Mythos comes arsenic questions equine astir whether the model's capabilities tin beryllium contained astatine all. Decrypt reported past week that researchers astatine Vidoc Security reproduced respective of Mythos's astir alarming cybersecurity findings utilizing publically disposable models—including OpenAI’s GPT-5.4 and Anthropic’s ain Claude Opus 4.6—without immoderate peculiar entree to Mythos itself.

Anthropic did not instantly respond to a petition for remark by Decrypt.