The Pentagon has switched on ChatGPT Mil and Grok for Government inside GenAI.mil, opening generative AI to more than 3 million defense personnel. Both join Google Gemini, the platform's first model. Claude is still missing, though a court struck down the label that excluded Anthropic on 28 August.
Key takeaways
- ChatGPT Mil (OpenAI) and Grok for Government (Starshield AI) went live on GenAI.mil
- Access covers 3 million-plus troops, civilian staff and contractors
- The portal has 1.7 million-plus unique users
- Both models cleared Impact Level 5 certification
- Anthropic's Claude is still unavailable
Three models, one portal
GenAI.mil is the Pentagon's centralized entry point to generative AI, launched last year. Gemini came first. Two commercial models adapted for military use now sit next to it.
ChatGPT Mil covers chat, files, projects and custom GPTs?Custom GPT: A version of ChatGPT configured for one specific task, with its own instructions, reference documents and tools, reusable across a team.. The department aims it at routine unclassified document work: administration, logistics, planning, policy. Grok for Government, delivered by SpaceX unit Starshield AI, gets a broader pitch: “immediate productivity gains, stronger knowledge continuity and more secure and efficient collaboration”.
| Vendor | Product | Status |
|---|---|---|
| Gemini | first on the platform | |
| OpenAI | ChatGPT Mil | live 31 Aug 2026 |
| Starshield AI (SpaceX) | Grok for Government | live 31 Aug 2026 |
| Anthropic | Claude | unavailable |
The portal logged over 1.7 million unique users before ChatGPT and Grok arrived.
Both hold Impact Level 5?Impact Level 5: A U.S. Department of Defense cloud accreditation tier (Cloud Computing SRG) for sensitive unclassified data. It requires isolated compute infrastructure and encryption of data at rest. certification, required for sensitive unclassified data.
The empty seat where Claude should be
Anthropic's absence is no accident. The Trump administration labeled the company a “supply-chain risk?Supply-chain risk: A formal administrative label marking a vendor as a threat to the department's supply chain. In practice it shuts the vendor out of Pentagon contracts and systems.” after it refused the Pentagon unrestricted access, insisting on guardrails against mass surveillance and autonomous weapons.
On 28 August, U.S. District Judge Rita Lin ruled the label unlawful, calling Defense Secretary Pete Hegseth's decision retaliation under the First Amendment, arbitrary, and a breach of due process.
The empty invocation of national security is not a blank check to punish and retaliate against government critics.
Rita Lin, U.S. District Judge, in the ruling.
Anthropic filed two March complaints, in California and Washington, D.C. The first ended with the August ruling, the second is still open.
Why it matters
The Pentagon is building an intermediary layer between commercial AI labs and users in uniform, and it alone decides whose model belongs there. The Anthropic case shows selection can turn on politics rather than technical merit.
For vendors, defending guardrails may cost access to one of the largest AI markets. Scale matters too: 1.7 million users ranks among the largest generative AI rollouts in any public administration.
What's next
- Anthropic's second complaint, filed in Washington, D.C. in March, still awaits a decision
- The August ruling does not compel the department to add Claude — that call is the Pentagon's
- GPT-5.6 Terra is coming to ChatGPT Mil
