Pentagon signs AI deals with OpenAI

The Pentagon just moved commercial AI into some of its most sensitive computing environments. That is the real story here — not a flashy new weapon, but the plumbing that determines which models can touch secret and top-secret work. On May 1, the Defense Department said OpenAI, Google, Microsoft, Nvidia, Amazon Web Services, SpaceX, Reflection, and later Oracle can deploy frontier AI capabilities on classified networks. Anthropic was not on the list. ### What did the Pentagon actually sign? These are formal agreements that let approved companies bring AI tools into the Pentagon’s Impact Level 6 and Impact Level 7 environments. In plain English, that means systems used for secret and highly sensitive national-security warfighting, intelligence, and enterprise tasks. ### Why do IL6 and IL7 matter? Because this is the hard part. Plenty of companies can demo a chatbot on an unclassified laptop. Far fewer can clear the security, hosting, and compliance hurdles needed to operate around classified data. IL6 is the standard for classified cloud services to become part of everyday defense workflows. ### So is this about one company winning? Not really. The Pentagon is doing the opposite of a single-vendor bet. Its own language says the goal is to avoid AI vendor lock-in and keep long-term flexibility for the Joint Force. That helps explain the broad roster — cloud giants, model labs, a chip company, a startup, and SpaceX. Basically, the department wants a menu, not a monopoly. ### Why is Anthropic missing? That is the sharpest part of the story. Anthropic and the Pentagon have been fighting over use restrictions — especially around autonomous weapons and surveillance of Americans. The Pentagon wanted terms allowing use for any lawful purpose as part of a broader government offloading push. ### Does Anthropic’s absence mean it is out everywhere? Not exactly. One report says Anthropic’s Claude had already appeared on classified networks through Palantir’s Maven toolkit, which shows how messy this ecosystem is. A company can be politically out with one vendor; being excluded from the official roster is still a real setback. It signals who defense leaders want as default suppliers right now. ### Why include Nvidia and SpaceX? Because military AI is not just about models. Nvidia matters because chips and AI infrastructure shape what can run at scale. SpaceX matters because defense computing increasingly connects to communications, edge deployment, and broader battle roles alongside much larger firms. ### What changed this week? The Pentagon had already been building GenAI.mil, its secure AI platform, and some companies were already working with defense systems. But this week made the vendor map much clearer. Google had recently rolled out Gemini 3.1 Pro on GenAI.mil, a shortlist. ### Bottom line? This is a market signal dressed up as a security announcement. The Pentagon is telling the industry which companies it trusts to handle classified AI now — and just as importantly, which one it does not.