The National Security Agency is reportedly using Anthropic's Mythos Preview, signaling expanding adoption within the agency despite a Pentagon designation of Anthropic as a supply chain risk.

Axios reports and multiple sources indicate NSA’s rollout of Mythos Preview comes amid ongoing disputes between Anthropic and the Pentagon over model use and government tools.

Mythos disclosures included patched vulnerabilities, with some findings shared as cryptographic hashes before full disclosure after fixes.

A recent meeting described as productive discussed next steps for Mythos use across departments beyond the Pentagon, though both Anthropic and the Pentagon declined comment.

Access to Mythos is restricted for now, with long-term plans aiming for a safer, scalable release via protective mechanisms and testing on a lower-risk model like Claude Opus.

Security concerns persist about Mythos’ advanced coding and autonomous capabilities potentially elevating cyberattack risks and deployment challenges.

Experts warn the model could enable cyber exploits by identifying vulnerabilities and proposing attack methods, underscoring urgent defense considerations.

The development follows discussions between the Trump administration and Anthropic’s CEO on collaboration, marking the first talks since tensions earlier this year over model deployment with the Pentagon.

UK counterparts are testing Mythos access through the AI Security Institute, illustrating international use and evaluation.

Anthropic restricts Mythos to about 40 organizations for responsible-use reasons, with publicized users limited; the NSA is among unnamed users per some sources.

A policy clash exists between defense officials seeking broad lawful-use access and Anthropic advocating limits on mass surveillance and autonomous weaponizable applications.

Axios notes growing NSA integration of Mythos, while independent verification attempts with Reuters faced unresponsiveness from involved parties.