NSA Reportedly Using Anthropic's Mythos Model Despite Pentagon Conflict

The U.S. National Security Agency (NSA) is among the undisclosed recipients of Anthropic's Claude Mythos model, according to Axios, citing internal sources. The revelation comes despite an active presidential directive from Donald Trump ordering all federal agencies to completely abandon Anthropic's AI technologies within six months.

The Anthropic–Pentagon Fallout

The conflict traces back to Anthropic's refusal to allow its AI models to be used for mass surveillance of U.S. citizens or the development of autonomous weapons. These disagreements with the Pentagon escalated to the point where President Donald Trump ordered every federal agency to fully phase out the startup's tools within a six-month window.

Yet despite that mandate, the NSA has secured access to Mythos — Anthropic's most powerful neural network to date.

Why This Matters

The situation exposes a stark contradiction within U.S. national security institutions. Military agencies are simultaneously expanding their use of Anthropic's tools while arguing in court that those very technologies pose a threat to national security. This paradox raises serious questions about the enforceability of the White House ban and suggests that Claude Mythos's capabilities proved too valuable to abandon.

Claude Mythos and Project Glasswing

Anthropic developed Claude Mythos in April but decided against a public release due to severe security risks. During just a few weeks of testing, the neural network discovered thousands of zero-day vulnerabilities across major operating systems and web browsers.

Instead of making the model publicly available, Anthropic launched Project Glasswing — a controlled testing initiative involving major industry players. Participants include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and others.

How exactly the NSA is deploying Mythos remains unknown. Other Glasswing participants are reportedly using the model to scan their own environments for vulnerabilities.

Signs of a Thaw

There are indications that Anthropic's relationship with Washington may be warming. On April 18, CEO Dario Amodei met with administration officials to discuss potential collaboration. White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent were among those present at the meeting.

Separately, Anthropic recently launched Claude Managed Agents — an environment designed to run complex, long-running agentic tasks — further expanding its AI product ecosystem.

Original post on ForkLog AI: Telegram channel