Anthropic's Claude had been integrated into the Pentagon's Maven Smart System and deployed on classified networks since July 2025. With LLMs on board, the US military could process five times as many targets per day. But as the Pentagon tried to renegotiate the terms of that arrangement, it ran into Anthropic's red lines: no fully autonomous weapons, and no mass domestic surveillance.

Bloomberg reporter Katrina Manson, author of a new book on Project Maven, breaks down what actually happened when Anthropic and the Pentagon fell out over the terms of Claude's military deployment. Manson says the dispute had a performative dimension on all sides, but it highlighted a real tension around the creation of guardrail for military use of LLMs.

As the Pentagon pushed to loosen those restrictions, Anthropic drew lines it wasn't willing to cross. Anthropic had separately applied to use LLMs in voice-controlled autonomous drone swarming technology, suggesting it was willing to go quite far, just not all the way to full autonomy.