The U.S. Military Now Has Its Own Versions of ChatGPT and Grok

More than 1.7 million military and civilian Pentagon personnel are already using a generative AI platform most Americans have never heard of. On August 31, 2026, the Department of Defense added two more models to GenAI.mil — ChatGPT Mil from OpenAI and Grok for Government from SpaceX’s Starshield AI division — joining Google Gemini on what is now the military’s official AI hub for 3 million total staff. Every major frontier model is represented. Except one.

Two New Tools, One Big Platform

GenAI.mil runs at Impact Level 5 (IL5), cleared for Controlled Unclassified Information — meaning sensitive but not classified data stays inside government infrastructure, isolated from consumer training systems.

Here’s what the two new additions actually do:

• ChatGPT Mil handles document-heavy unclassified work — planning, policy drafting, logistics paperwork
• Grok for Government offers adaptive reasoning modes (Auto, Fast, Expert) for more complex analysis
• Both run on isolated government cloud infrastructure, separate from public versions
• Grok’s “playbooks” capture institutional knowledge for reuse across missions — essentially a searchable unit memory
• Pentagon-stated use cases: acquisition market research, supply-chain management, staff work

The One Company That Said No

Claude is now the only major model missing from GenAI.mil — and the story of how it got there involves a federal court, a blacklisting, and a constitutional fight Anthropic is currently winning.

Contract negotiations between Anthropic and the DoD reportedly collapsed over a phrase: “any lawful purpose.” Anthropic refused language that could permit domestic mass surveillance or autonomous weapons use. Defense Secretary Pete Hegseth responded by designating Anthropic a “supply chain risk” under 10 U.S.C. § 3252 — a procurement statute ordinarily aimed at foreign threats to military systems, not AI companies headquartered in San Francisco.

“Classic illegal First Amendment retaliation.” — U.S. District Judge Rita Lin, describing the Pentagon’s blacklisting of Anthropic, according to Politico and CNBC

Judge Lin issued a preliminary injunction in March 2026. By late August, a second ruling held the designation constituted unlawful retaliation under both the First and Fifth Amendments. A second case in the D.C. Circuit remains ongoing, according to Reuters and Lawfare Media.

What Gets You Past the Velvet Rope

Getting onto GenAI.mil, it turns out, is less a procurement process and more a values audition.

The AI industry’s vendor relationships with the Pentagon now work like this: agree to broad usage rights, you’re in; push back on surveillance and autonomous weapons, and you risk being labeled a national security threat. That’s the selection mechanism behind a 3-million-person government contract. Agree to broad usage terms and your model ships to warfighters. Refuse, and your company ends up in federal court — though, notably, winning.

If courts keep blocking retaliatory blacklisting, more AI companies may choose to build usage restrictions directly into contracts, reshaping how defense procurement gets written. What’s actually at stake here extends well beyond logistics software and supply-chain chatbots. Two federal rulings now say the government cannot punish a company for its public stance on AI safety — and the AI industry is watching every word of the next contract.