The NSA Is Reportedly Using Anthropic's Mythos for Cyberattacks

There is a Mythos and NSA story, and it is a strange one. According to a Financial Times report, the National Security Agency is using Anthropic's most powerful cyber model, Claude Mythos, for offensive cyber operations, with about half a dozen Anthropic engineers embedded inside the agency. The twist is that the rest of the US government has been trying to keep Anthropic out.

This is the same model at the center of the export fight. Mythos is the frontier system Anthropic built for cybersecurity, and it is good enough that the company called it too dangerous to release widely. In testing it autonomously found thousands of unpatched zero-day vulnerabilities across every major operating system and browser, including a 17-year-old remote-code flaw in FreeBSD. That power is exactly why the government wants it, and why it is so sensitive.

The contradiction is the story. The Pentagon labeled Anthropic a supply-chain risk to national security, and the White House barred the company from federal use and slapped export controls on its newest models. Yet the NSA, reportedly, is doing the opposite, deploying Mythos for offensive operations and having Anthropic staff teach its officials how to build custom tools with it. One arm of the state shuts Anthropic out while another quietly brings it in.

Anthropic is private, so there is no stock to trade, but the read for the AI sector is real. It shows that the most capable cyber AI is now a national-security asset, wanted by the same government that fears it in the wrong hands. For Anthropic, weeks before a planned IPO, being indispensable to the NSA is a strange kind of validation, and a reputational risk, since a safety-first brand working on offensive cyber operations cuts against its whole image.

The bigger issue is what this means for AI and cyber war. A model that can find and exploit zero-days faster than almost any human changes the offense-defense balance, and putting it inside an intelligence agency is a serious escalation. It also exposes how messy US AI policy has become, with export bans, a Pentagon blacklist, and an NSA deployment all aimed at the same company at the same time. Nobody seems fully in charge of the policy.

So your instinct was right, there is a Mythos and NSA story, just not a breach. It is the NSA using Anthropic's most dangerous model while the rest of Washington tries to lock it down. Banned by one hand, embraced by another. That contradiction is the real headline, and it is not resolved.