Pentagon adds Grok and ChatGPT to GenAI.mil; Grok for Government hits 1.7M users

The US Department of War announced on August 31 that it has expanded its enterprise generative-AI portal, GenAI.mil, to include xAI's Grok and OpenAI's ChatGPT alongside Google's Gemini, which was the first commercial model to go live on the secure platform. In parallel, the department launched Starshield AI, deploying xAI's Grok for Government to 1.7 million users — military members, civilian employees, and contractors — in what is being described as one of the largest single-model government deployments to date. OpenAI's offering ships as 'ChatGPT Mil' at IL5 accreditation.

The design intent is model choice: rather than standardize on one vendor, the Pentagon wants personnel to tap the distinct technical strengths of multiple top-tier LLMs for specialized missions. That mirrors the multi-model approach AWS is pushing in GovCloud, where Bedrock now exposes Claude, Llama, OpenAI, Grok and Nova to federal customers.

Competitively, the deals mark a striking acceleration of frontier-AI adoption inside defense, and a win for xAI in particular, which secures scale distribution to 1.7M seats. It also puts three rival labs — Google, OpenAI, xAI — inside the same government procurement envelope, intensifying head-to-head comparison on classified-adjacent workloads.

The caveats are governance-shaped: community discussion flags vendor lock-in, oversight, and the security implications of routing sensitive military tasks through commercial models. With accreditation levels like IL5 in play, the open question is how the department audits model behavior, data handling, and hallucination risk at 1.7M-user scale — and whether the multi-vendor model choice actually reduces or compounds oversight complexity.