- The NSA is reportedly using Mythos Preview, Anthropic’s cybersecurity AI model that the company withheld from public release due to its capacity for offensive cyberattacks.
- Anthropic capped Mythos access at approximately 40 organizations; the NSA is among the undisclosed recipients and is using the model for vulnerability scanning.
- The Department of Defense — the NSA’s parent agency — labeled Anthropic a “supply chain risk” weeks before NSA access was confirmed, after Anthropic refused Pentagon officials unrestricted access to the model.
- The UK’s AI Security Institute has also confirmed it has access to Mythos, making it one of roughly a dozen organizations Anthropic has publicly named.
What Happened
The National Security Agency is reportedly using Mythos Preview, Anthropic’s restricted cybersecurity AI model, according to a TechCrunch report published April 20, 2026, citing Axios. Anthropic announced Mythos earlier this month as a frontier model built for cybersecurity tasks but immediately withheld it from general release, stating the model was too capable of enabling offensive cyberattacks. The NSA is said to be deploying Mythos specifically to scan environments for exploitable vulnerabilities.
Why It Matters
The reported NSA deployment arrives weeks after the Department of Defense labeled Anthropic a “supply chain risk” — a designation that originated when Anthropic refused to make Claude available for mass domestic surveillance and autonomous weapons development, and declined to grant Pentagon officials unrestricted access to Mythos’s full capabilities. The Pentagon has since argued in court that Anthropic’s tools can threaten national security. The NSA’s separate, reported access to the same model means different arms of the U.S. government are simultaneously litigating against Anthropic and operationally depending on its technology.
Technical Details
Anthropic described Mythos Preview as a frontier-class model optimized specifically for cybersecurity applications — a narrower scope than any prior Claude release. The company determined the model crossed a threshold for offensive cyberattack enablement that it has not previously invoked to restrict a model’s release. Total access was capped at approximately 40 organizations, of which Anthropic has publicly identified roughly a dozen. The NSA’s reported use case is defensive: automated scanning of environments to identify exploitable vulnerabilities before adversaries can reach them.
Who’s Affected
The NSA and allied intelligence agencies gain operational access to an AI model that Anthropic itself assessed as too capable for public release, establishing a channel for government use of AI tools withheld from private and commercial users. The UK’s AI Security Institute confirmed it is among the disclosed Mythos recipients, indicating allied government bodies are coordinating access. The approximately 28 undisclosed partner organizations remain unidentified, limiting public oversight of how the model is being used across those deployments.
What’s Next
Anthropic’s relationship with the Trump administration appears to be stabilizing. On April 18, 2026, Anthropic CEO Dario Amodei met with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent; the White House described the meeting as “productive.” The Pentagon’s active legal dispute with Anthropic over the supply chain risk designation has not been publicly resolved, and Anthropic declined to comment on the NSA’s reported Mythos access.
