NSA Using Anthropic's Restricted Mythos Model Despite Pentagon 'Supply-Chain Risk' Designation
Summary
- • NSA reportedly using Mythos Preview, Anthropic's restricted cybersecurity AI model withheld from public release
- • Pentagon simultaneously labels Anthropic a supply-chain risk while its own intelligence arm deploys the tool
- • U.K. AI Security Institute also confirmed as a Mythos access recipient
- • Anthropic CEO met with senior White House officials; relationship described as productive
Details
NSA reportedly using Mythos Preview for vulnerability scanning
The NSA appears on the undisclosed portion of Anthropic's ~40-organization access list, using Mythos primarily to scan environments for exploitable vulnerabilities — a defensive security application of a model Anthropic deemed too dangerous for public release.
Mythos Preview: frontier cybersecurity model restricted to ~40 organizations
Announced earlier in April 2026, Mythos was withheld from public release because Anthropic judged it capable enough of offensive cyberattacks to pose unacceptable public risk. Only about a dozen of the ~40 approved recipients have been publicly named.
Pentagon designated Anthropic a 'supply-chain risk' over access dispute
The DoD applied the supply-chain risk label after Anthropic refused to provide Pentagon officials unrestricted access to Claude's full capabilities — specifically for mass domestic surveillance and autonomous weapons development.
U.S. government taking contradictory stances toward Anthropic simultaneously
The DoD is pursuing legal and procurement action against Anthropic while the NSA — a DoD component — actively deploys Anthropic's most sensitive restricted model, suggesting either poor inter-agency coordination or a deliberately fragmented approach to AI procurement.
U.K. AI Security Institute confirmed as a Mythos access recipient
The U.K.'s AI Security Institute has publicly confirmed it holds access to Mythos, making it one of the few publicly named organizations alongside the now-implied NSA.
Amodei met with White House Chief of Staff and Treasury Secretary
Dario Amodei met with Susie Wiles and Scott Bessent. The White House called the meeting productive, signaling the Anthropic-administration relationship may be improving at the executive level even as the DoD dispute continues.
Pentagon simultaneously arguing in court that Anthropic's tools threaten national security
Even as the NSA deploys Mythos, the DoD is pursuing legal arguments that Anthropic's AI poses national security risks — a contradiction occurring within the same parent organization.
Industry Update = operational developments; Product Launch = new model details; Policy = government designations; Insight = analytical observation; Partnership = confirmed access; Strategy = executive-level moves; Legal = ongoing litigation
What This Means
The U.S. government's relationship with Anthropic is fractured along agency lines: the Pentagon is pursuing a supply-chain risk designation while one of its own intelligence arms quietly deploys the company's most restricted AI model. This exposes a broader tension in how governments regulate frontier AI — blanket risk labels and agency-level procurement decisions can diverge sharply, creating incoherent policy in practice. For Anthropic, the situation is double-edged: NSA adoption legitimizes Mythos as a serious national security tool, but the simultaneous Pentagon dispute signals that its refusal to enable certain military use cases carries real institutional costs.
Sentiment
Pragmatic support for NSA use amid government tensions, dismissing Pentagon label as symbolic
“the geopolitical/security imperative outweighs everything else, this is a perfect encapsulation... Petty squabbles will always be triumphed by national security imperatives”
“The same week the Pentagon declared Anthropic a "supply chain risk." The same week the NSA confirmed it was using Mythos anyway. The same week the White House met with Dario Amodei to find a path forward.”
“so, safe to say the Pentagon "supply chain risk" thing was a juicy nothingburger?”
“The Pentagon labeled Anthropic a supply-chain risk. Yet the NSA is reportedly using the company's restricted Mythos model for vulnerability scanning anyway. The military is busy arguing in court that these tools are a national security threat while quietly using the most dangerous versions of them.”
Split
Security pragmatists (national security needs first, ~70%) vs critics highlighting hypocrisy (~30%).
Sources
- Trump officials may be encouraging banks to test Anthropic’s Mythos modelTechCrunch
- Anthropic co-founder confirms the company briefed the Trump administration on MythosTechCrunch
- NSA spies are reportedly using Anthropic’s Mythos, despite Pentagon feudTechCrunch
- Anthropic Embeds Engineers in the NSA to Deploy Mythos for Offensive Cyber (3 minute read)Implicator