The NSA is using Anthropic's Mythos platform for cyber operations. It's also labelled it as a supply chain risk after Trump officials moved to block defense contractors from its models. Axios broke the story, Reuters confirmed it, and the contradiction is hard to miss.
Either the risk designation is overstated, or the NSA is using a dangerous tool because it's too good to pass up.
Hacker News commenters spotted this immediately. One suggested Anthropic is creating artificial scarcity around Mythos and its companion platform Glasswing, forcing agencies into a corner: deny access to what's been called "the most advanced hacking tool on the planet," or admit the blacklist doesn't mean much. Whether that's intentional corporate strategy or just the natural result of building superior tech, it's working.
Technical details on Mythos remain thin. It likely builds on Anthropic's Claude model family, fine-tuned for cybersecurity and intelligence work. Glasswing appears to be the orchestration layer handling deployment. Both probably use Anthropic's Constitutional AI framework, adapted for offensive operations rather than chatbot safety.
The NSA wants Mythos. Anthropic eyes a classified AI deal with US intelligence regarding the tool. The blacklist says they shouldn't have it, but when a tool can supposedly outperform every other hacking platform on the planet, blacklists become suggestions, not rules.