Pentagon Puts ChatGPT and Grok in Front of 3M Troops

The Pentagon on Monday switched on two more frontier chatbots inside GenAI.mil, the Defense Department's centralized generative-AI portal: OpenAI's ChatGPT Mil and Starshield AI's Grok for Government. Both join Google Cloud's Gemini for Government, which has had the portal to itself since it launched in December 2025. The department says GenAI.mil has now onboarded more than 1.7 million unique users out of roughly 3 million eligible personnel.

The point of the portal is data containment. Rather than have service members paste logistics spreadsheets and draft policy memos into consumer apps, the department routes them through military-tailored builds that are exempt from the telemetry and training-data collection that is hard to switch off in commercial products. Both new models cleared Impact Level 5 (IL5) authorization, the Defense Department tier covering non-public, sensitive unclassified information — which means controlled unclassified information can go in, but classified work still cannot.

ChatGPT Mil ships with chat, file handling, month-long persistent projects and custom GPTs, aimed squarely at document-heavy routine work: travel forms, contracting opportunities, planning, policy and administration. It currently serves GPT-5.4 Terra, with OpenAI saying GPT-5.6 Terra is queued behind it, and includes an offline search capability that returns detailed citations. Grok for Government leans on its reasoning ladder — Auto, Fast and Expert inference modes — plus persistent projects and reusable playbooks, pitched at market research, supply-chain analysis and faster mission execution. It reaches the portal through Starshield AI, the xAI/SpaceX defense vehicle.

One name is conspicuously absent. Anthropic's Claude is not on GenAI.mil, and its absence is not a procurement backlog. The administration designated the company a "supply-chain risk" roughly six months ago after it declined to grant the department unrestricted access and insisted on keeping its usage guardrails in place. Anthropic sued. Last week a federal court ruled the Pentagon's designation illegal and baseless — and Monday's launch went ahead without Claude regardless.

That gap has been visible from the inside for months. In May, BitsMinds reported on the 25 Pentagon "power users" running an informal bake-off between OpenAI, Google and Grok on one side and Claude on the other, and on the eight-vendor round of AI agreements that cut Anthropic out. What changed on Monday is scale: the bake-off is over, and two of the three winners are now a login away for anyone with a common access card.

A War Department official framed the expansion in capability terms, saying access to a diverse suite of AI tools "will give warfighters the tools they need to act with confidence." The harder question the portal now raises is a governance one. Three vendors, three different safety postures and three different reasoning stacks are being handed to 3 million people for unclassified work, while the one lab that made guardrails a condition of the deal sits outside the building with a court ruling in its favor.