Pentagon Opens ChatGPT and Grok to 3 Million Staff, Claude Absent

The world's largest employer has started handing out AI licenses at scale. On August 31 the US Department of Defense, now styled the Department of War, added OpenAI's ChatGPT Mil and Grok for Government, supplied through SpaceX's Starshield AI, to its GenAI.mil portal. The portal launched with a military build of Google's Gemini; of the department's 3 million military and civilian personnel, 1.7 million are already onboarded. This is the largest enterprise AI deployment inside a single organization anywhere.

The real significance is that the public sector has just retired the sentence "we handle sensitive data, so we can't use AI." The Pentagon did not do it by opening the consumer products. It bought the same models installed inside its own secure environment.

The terms of the deal

The numbers, per Fortune: OpenAI and SpaceXAI each hold contracts capped at $200 million; Google and Anthropic have older contracts whose values were not disclosed. The technical envelope is Impact Level 5 in the US government cloud taxonomy, the tier that permits Controlled Unclassified Information. Data stays inside the environment and is not used to train the models. In the Pentagon's words, ChatGPT Mil "brings a familiar commercial experience into the Department's secure environment, tailored to warfighter needs."

The listed uses are ordinary office work: administrative writing, logistics planning, policy drafting, market research, supply-chain analysis. The portal is aimed at millions of desk jobs, not weapons systems. Other contract holders in the ecosystem include AWS, Microsoft, Nvidia and Reflection AI.

Why Anthropic is missing

Claude is not on the list, and that is not an accident. After Anthropic refused the Pentagon's demand for unrestricted use and insisted on safety guardrails, the administration designated the company a "supply-chain risk," and in February federal agencies were told to stop using Anthropic technology. Anthropic sued, and last week a federal judge ruled the blacklist unlawful. That the portal's latest expansion still excludes Anthropic shows the ruling has not yet reached the procurement floor.

Enterprise buyers should read that as a signal too. Model selection is no longer purely a technical choice; a vendor's relationship with government now lands in the contract.

The lesson for regulated organizations

Banks, hospitals and public bodies everywhere postpone AI projects for the same reason: data leaving the perimeter, privacy law, auditability. The Pentagon example shows the shape of the answer: same model, different installation. Data does not leave, it is not used for training, and the access level is written into the contract.

The equivalent for a regulated company is running frontier models in a sovereign cloud region or a dedicated environment. Those options are expensive, but they exist; the bottleneck is procurement and contract literacy, not technology. Our short assessment: an organization that says "we can't use AI" today usually means "we can't use the consumer version." The gap between those two sentences decides whether the project happens.

Sources: Fortune, TechCrunch