According to Axios, which cited two sources, the National Security Agency is reportedly employing Anthropic’s Claude Mythos Preview within its classified networks. This development is notable as it occurs despite the Department of Defense, under which the NSA operates, having previously labeled Anthropic a supply-chain risk in March and currently contesting the company in federal court.
Anthropic has restricted access to Claude Mythos due to concerns about offensive security risks. The model was found capable of identifying significant vulnerabilities in all major operating systems and web browsers, leading Anthropic to limit its release. Most organizations accessing the tool, including Microsoft, Google, Apple, Amazon Web Services, JPMorgan Chase, and Nvidia, are using it defensively to identify infrastructure weaknesses before adversaries.
The NSA’s use of Mythos is less defined but reportedly extends beyond defensive applications within the intelligence community, as per a third source. The Pentagon’s antagonism toward Anthropic stems from failed negotiations in July 2025, when the two parties agreed to make Claude the first frontier AI model approved for classified networks. Talks deteriorated after the Pentagon demanded unrestricted military access, which Anthropic denied due to its policies against autonomous weapons and domestic mass surveillance.
Following unsuccessful renegotiations, Defense Secretary Pete Hegseth declared Anthropic a supply-chain risk in late February, an unprecedented action against an American firm. While a California court initially blocked this move, a D.C. appeals court later dismissed Anthropic’s request to suspend the blacklisting during ongoing litigation.
Meanwhile, the administration is exploring other avenues. On April 17, Anthropic’s CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent, which Reuters described as ‘productive’. The meeting focused on collaboration opportunities and addressing challenges related to scaling this technology.
President Trump expressed ignorance about Amodei’s visit to the White House, despite previously ordering a halt on using Anthropic’s models. Meanwhile, Federal Reserve Chair Jerome Powell and Treasury Secretary Bessent are urging major bank CEOs to prepare for security threats by testing Mythos. An administration source disclosed that every federal agency except the Defense Department seeks access to Anthropic’s tools.
The NSA’s reported use of Mythos raises concerns about controlling such capabilities. Decrypt recently revealed that researchers at Vidoc Security replicated some of Mythos’s critical cybersecurity findings using publicly available models like OpenAI’s GPT-5.4 and Anthropic’s Claude Opus 4.6, without direct access to Mythos.
Anthropic has not yet responded to Decrypt’s request for comment.