Two sources tell Axios the NSA has access to Mythos Preview, Anthropic's most restricted model. Only about 40 organizations can use it. Meanwhile, the Department of Defense insists Anthropic is a national security threat and argues that in court. The same Pentagon keeps expanding internal use of those tools.
Mythos focuses on offensive cybersecurity. Fine-tuned to find and exploit vulnerabilities rather than hold conversations, it lets organizations scan their own systems for security holes. The model can execute multi-step penetration tests on its own, catching zero-day flaws in complex code.
The blacklist stems from a contract dispute. The Pentagon demanded Claude be available for "all lawful purposes," meaning mass domestic surveillance and autonomous weapons. Anthropic said no. Some Defense officials called that refusal proof the company can't be trusted when it matters. CEO Dario Amodei met with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent on Friday to work through the mess. Both sides described the meeting as productive. Next steps will likely focus on how non-Pentagon agencies get access to Mythos. The intelligence community wants access to the model. The blacklist isn't stopping them.