Spy Games: The NSA Snags Anthropic’s Restricted AI Despite Pentagon Beef

The National Security Agency is currently making headlines for using a tool that the public cannot touch. Reports suggest that the NSA is working with Mythos Preview, a highly restricted AI model from Anthropic. This news comes at a strange time. Just a few weeks ago, the Department of Defense labeled Anthropic as a potential supply chain risk. The Pentagon was angry because Anthropic refused to give officials total access to everything the model can do. Even with that drama in the background, it looks like the spies at the NSA found a way to get what they wanted.

Anthropic first announced Mythos earlier this month. They built it as a frontier model specifically for cybersecurity tasks. However, the company made a surprising choice. They claimed the model was actually too good at finding ways to attack other systems. Because they feared people would use it for offensive cyberwarfare, they decided to keep it away from the general public. Instead, they gave access to only about 40 organizations. The NSA appears to be one of those secret users.

Cybersecurity or Cyber-Attacks?

The NSA is likely using Mythos to scan digital environments. They want to find exploitable holes in software before hackers do. This sounds like a standard security move, but the context is messy. The U.S. military is currently arguing in court that tools like this could actually threaten national security if they fall into the wrong hands. The original fight with the Pentagon started because Anthropic did not want their AI used for mass surveillance or to build autonomous weapons.

By keeping the model restricted, Anthropic tried to stay on the side of safety. But when the most powerful spy agency in the world becomes your primary user, those safety lines get blurry. The U.K. AI Security Institute also confirmed that they have access to the model. This shows that while the average person cannot use Mythos, the most powerful governments on earth are already putting it to work.

A Political Thaw

The relationship between Anthropic and the government seems to be changing fast. Last Friday, Anthropic CEO Dario Amodei met with high-level White House officials, including the chief of staff and the Secretary of the Treasury. The White House called the meeting productive, which suggests that the frost between the tech firm and the Trump administration is starting to melt.

This partnership is a big deal for the AI industry. It shows that the most advanced tools are no longer just for writing emails or making art. They are becoming central pieces of national defense and intelligence. As Anthropic moves closer to the government, they have to balance their promise of “safe AI” with the demands of agencies that specialize in secrets and surveillance. For now, the NSA has a powerful new ally in its digital basement, even if the rest of the Pentagon is still skeptical.