Anthropic

Anthropic, the San Francisco lab founded in 2021 by a group of former OpenAI staff led by Dario Amodei , built its reputation on the Claude family of models and a public emphasis on AI safety. That safety framing sits alongside a growing line of work with the US national-security establishment.

The route into defence ran first through Palantir . In November 2024 Anthropic, Palantir and Amazon Web Services announced a partnership to make Claude models available to US defence and intelligence agencies through Palantir’s AI Platform running on AWS. The deployment was accredited at a level permitting work with classified data up to Secret, and Palantir described itself as the first industry partner to bring Claude into classified environments — a distinct path from selling a chatbot, instead embedding the models inside a platform already trusted by the agencies.

In June 2025 Anthropic introduced “Claude Gov,” a set of models built specifically for national-security customers and, the company said, already in use by agencies at the highest levels of government, with access restricted to classified settings. The models are tuned to handle classified material more readily and to better parse defence and intelligence documents. A month later the Pentagon’s Chief Digital and Artificial Intelligence Office awarded Anthropic a two-year prototype agreement with a $200 million ceiling — one of four near-identical deals signed the same day with Anthropic, OpenAI , Google and xAI to prototype frontier and agentic AI for the Department of Defense. Separately, a 2025 General Services Administration arrangement offered Claude to all three branches of government for a nominal fee. For a company that markets itself on caution, the trajectory is notable: in little over a year, a lab built around AI-safety research became one of the handful of frontier providers cleared to work inside America’s classified national-security systems.