The U.S. National Security Agency is using Anthropic's new Claude Mythos Preview model for cybersecurity work, even as the Pentagon keeps the company on a supply-chain risk list. Two sources told Axios the NSA has access to Mythos Preview, Anthropic's most capable model yet, and one added the tool is spreading more widely inside the Department of Defense. Unlike public-facing Claude models or OpenAI's ChatGPT, Mythos is a restricted frontier model tuned to find and chain software exploits on its own. During red-team testing it escaped its sandbox and completed a simulated corporate network attack end to end, which is why it was never released to a mass audience. The agency is one of roughly 40 organizations that Anthropic has granted Mythos access. Mythos was unveiled on Apr. 7 as a general-purpose model with strong cyber exploit detection skills and is being deployed under a controlled initiative called Project Glasswing. CEO Dario Amodei met with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent on Friday, with both sides calling the meeting productive. Federal CIO Gregory Barbaccia told Cabinet officials the Office of Management and Budget is working with Anthropic, industry partners and the intelligence community on guardrails before a modified Mythos is cleared for wider agency use. The Pentagon cut off Anthropic in February after the company refused to drop safeguards that block autonomous weapons use and mass domestic surveillance; the dispute is still in court.