NSA Is Utilizing Anthropic's Highly effective Claude Mythos AI as CEO Meets With White Home: Report – Decrypt

The Nationwide Safety Company is operating Anthropic’s Claude Mythos Preview inside its categorized networks, in response to two sources cited by Axios—a stunning growth provided that the NSA falls underneath the Division of Protection, which declared Anthropic a supply-chain threat in March and is at present combating the corporate in federal courtroom.

Claude Mythos just isn’t a regular enterprise software. When Anthropic unveiled the mannequin earlier this month, it restricted entry to a handful of vetted organizations, arguing that the mannequin poses critical offensive safety dangers. Anthropic’s personal technical documentation discovered that Mythos was capable of establish crucial vulnerabilities in each broadly used working system and net browser. The corporate judged it too harmful for open launch.

Most organizations with entry are utilizing the mannequin defensively, scanning their very own infrastructure for weaknesses earlier than adversaries do. The initiative, branded Challenge Glasswing, contains Microsoft, Google, Apple, Amazon Internet Providers, JPMorgan Chase, and Nvidia. What the NSA is doing with Mythos is much less clear, although the company’s mission just isn’t purely defensive. A 3rd supply instructed Axios the mannequin is getting used extra broadly throughout the intelligence division.

The Pentagon’s hostility towards Anthropic traces to negotiations that went dangerous. In July 2025, the 2 sides signed an settlement making Claude the primary frontier AI mannequin cleared to be used on categorized networks. Talks soured when the Pentagon sought to renegotiate, demanding the army be allowed to make use of Claude “for all lawful functions” with out restriction. Anthropic refused, drawing two agency strains: no autonomous weapons, and no home mass surveillance.

When negotiations collapsed, Protection Secretary Pete Hegseth declared Anthropic a supply-chain threat in late February—an unprecedented designation, and the primary ever utilized to an American firm. A California federal decide blocked the transfer, however then a D.C. appeals courtroom denied Anthropic’s separate bid to halt the blacklisting whereas litigation performs out. The 2 sides stay in courtroom.

Whereas the authorized battle grinds on, the remainder of the administration is shifting in a unique course. On April 17, Anthropic CEO Dario Amodei met with White Home Chief of Employees Susie Wiles and Treasury Secretary Scott Bessent. Anthropic described the session as “productive”, Reuters reported. The White Home stated the events “mentioned alternatives for collaboration, in addition to shared approaches and protocols to deal with the challenges related to scaling this expertise.”

President Trump, requested by reporters concerning the assembly, stated he had “no concept” Amodei had been on the White Home, after he beforehand ordered the administration to not use Anthropic’s “woke” fashions. Bessent and Federal Reserve Chair Jerome Powell have individually been encouraging main financial institution CEOs to check Mythos and be ready for safety threats, and an administration supply instructed Axios that each federal company besides the Protection Division desires entry to Anthropic’s instruments.

The NSA’s reported use of Mythos comes as questions mount about whether or not the mannequin’s capabilities may be contained in any respect. Decrypt reported final week that researchers at Vidoc Safety reproduced a number of of Mythos’s most alarming cybersecurity findings utilizing publicly accessible fashions—together with OpenAI’s GPT-5.4 and Anthropic’s personal Claude Opus 4.6—with none particular entry to Mythos itself.

Anthropic didn’t instantly reply to a request for remark by Decrypt.