NSA using Anthropic's Mythos despite blacklist
anthropic
| Source: HN | Original article
The National Security Agency has begun running Anthropic’s unreleased “Mythos Preview” model for cybersecurity and intelligence work, even though the Pentagon has formally labeled the San‑Francisco startup a “supply‑chain risk” and an executive order issued in February bars federal agencies from using Anthropic tools. Two senior sources told Axios that the NSA’s cyber‑defense teams are leveraging Mythos to parse threat‑intel feeds, automate vulnerability assessments and draft incident‑response briefings, despite the blacklist that was meant to keep the technology out of government hands.
The move matters because it pits two powerful parts of the U.S. security establishment against each other. The Department of Defense’s risk designation was intended to protect classified networks from potential backdoors or data‑exfiltration pathways embedded in third‑party AI models. By sidestepping that restriction, the NSA is effectively saying the operational benefits of Mythos outweigh the perceived supply‑chain hazards. The decision also raises questions about compliance with the February 27 executive order, which could trigger internal audits or congressional scrutiny.
As we reported on 19 April, finance ministers and top bankers were already voicing serious concerns about the model’s reliability and the misinformation surrounding its launch. The NSA’s adoption adds a new layer of urgency to those debates, highlighting how quickly high‑risk AI can slip into critical infrastructure despite formal bans.
Watch for a formal inquiry from the Office of the Director of National Intelligence, possible revisions to the Pentagon’s risk‑labeling framework, and Anthropic’s legal response to the agency’s use of an unreleased product. Equally important will be whether other intelligence or law‑enforcement bodies follow the NSA’s lead, potentially reshaping the balance between AI innovation and national‑security safeguards.
Sources
Back to AIPULSEN