NSA uses Anthropic's Mythos Preview for cyber defense despite Pentagon supply chain risk label

The National Security Agency is using Anthropic's Mythos Preview, the company's most advanced AI model, despite the Pentagon officially designating Anthropic as a "supply chain risk," according to reporting from Axios. This development underscores how immediate cybersecurity needs are superseding an ongoing dispute between the military and the artificial intelligence firm.

The conflict between the Pentagon and Anthropic began earlier this year during contract negotiations. The Department of Defense sought unrestricted access to Anthropic's Claude model for "all lawful purposes," but the company refused, citing concerns about mass domestic surveillance and autonomous weapons development. In response, the Pentagon labeled Anthropic a supply chain risk, and the dispute has escalated into litigation.

Despite this formal blacklisting, the urgency of the government's cybersecurity requirements appears to have won out. According to multiple sources, the NSA is employing Mythos primarily to scan its environments for exploitable vulnerabilities. Anthropic announced Mythos earlier in April as a frontier model designed specifically for cybersecurity tasks, but determined its capabilities were too dangerous for public release. The company has limited access to approximately 40 organizations, publicly naming only a dozen of them. The NSA appears to be among the undisclosed recipients.

Mythos represents a significant technological leap in offensive cyber capabilities. The model can autonomously find, analyze, and exploit software vulnerabilities at scale—in some cases more effectively than human experts. Anthropic has warned that the model is so powerful that even non-specialists could identify and exploit sophisticated security flaws. This dual-use potential is precisely why the company restricted its availability in the first place.

The NSA's access to Mythos also coincides with signs that Anthropic's relationship with the Trump administration is improving. Last week, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent, marking the first substantive discussion between Anthropic's leadership and the administration since the Pentagon dispute began. The United Kingdom's AI Security Institute has also confirmed access to Mythos, indicating the restricted model is being deployed by allied intelligence agencies as well.

The situation illustrates a fundamental tension in AI governance: the potential security benefits of advanced AI tools versus concerns about corporate oversight and control. The military's simultaneous pursuit of Anthropic's technology while arguing in court that it poses national security risks highlights the complexity of managing critical AI resources during periods of geopolitical competition and rapid technological change.