Anthropic has embedded engineers inside the NSA to deploy its Mythos AI model for offensive cyber operations, despite an ongoing legal battle with the Pentagon. The San Francisco-based company installed about half a dozen staff as forward-deployed engineers to guide the technology's use, according to sources familiar with the arrangement.

This partnership exists while Anthropic fights the Pentagon's "supply-chain risk" designation, a first for a US business. The AI firm tried to restrict government use of its Claude models for mass surveillance and lethal drones. Amid the dispute, Anthropic filed for an IPO valued at over $1tn and expanded Mythos distribution to 150 organizations across 15 countries.

The move comes as OpenAI released a similar model and Trump signed an executive order for AI security reviews. Anthropic's engineers may help infiltrate networks of nations like China or Iran, with one source noting that "the best way to build a good defence is to build a good attack." Both Anthropic and the Pentagon declined to comment on the collaboration.