The Pentagon's ChatGPT Is Live for 3 Million. Guess Who's Missing.

TL;DR

The Pentagon switched on ChatGPT Mil and Grok for Government on August 31, adding both to GenAI.mil, the department-wide generative AI portal that has run on Google's Gemini alone since December 2025. The portal already counts more than 1.7 million users, and OpenAI says its military build is engineered to scale to the full workforce of over 3 million military members, civilians, and contractors, per The Hill. Everything runs at Impact Level 5, which means sensitive-but-unclassified paperwork, not war plans. The loudest part of the announcement is an absence: Claude is still not on the portal, four days after a federal judge ruled the Pentagon's blacklisting of Anthropic unlawful.

Three chatbots and an org chart

GenAI.mil launched in December 2025 with exactly one model family, Gemini for Government, and the Pentagon has been promising a multi-vendor roster ever since. It now has one, per DefenseScoop: Gemini, ChatGPT Mil, and Grok for Government, the last sold through Starshield AI, the SpaceX-side vendor that now ships xAI's Grok to federal customers. Yes, the military now buys one of its chatbots from a rocket company.

The whole portal is accredited at Impact Level 5, the DoD cloud authorization tier for Controlled Unclassified Information. Impact levels work like a clearance ladder for software instead of people: IL5 clears a system to touch the sensitive-but-unclassified paperwork that makes up most of the department's day, while anything classified lives several rungs higher, in environments none of these tools are cleared for.

What 1.7 million users actually get

ChatGPT Mil is not consumer ChatGPT with a camouflage skin. Per Defense One, it runs in a separate environment from the commercial product, went through security testing that started in February 2026 (blowing past an early-July launch target), and was reviewed by both DISA and NSA. Data stays inside government systems and is exempt from consumer-side data collection.

The feature list reads like ChatGPT Enterprise translated into government: chat, file uploads, persistent projects, custom GPTs, and reusable "playbooks that capture and scale institutional knowledge," which is an ambitious claim to make about any software pointed at several decades of accumulated DoD PowerPoint. The Pentagon pitches it at document-heavy unclassified work: planning, policy, logistics, administration. DefenseScoop reports it currently serves GPT-5.4 Terra, with GPT-5.6 Terra coming soon.

Grok for Government gets pitched at more operational chores, "from market research analysis for acquisition professionals to supply chain management for logisticians," in the department's words. A defense official's division of labor, per Defense One: Gemini for search, ChatGPT for text-heavy work.

The empty chair

The model missing from the roster is the story under the story. Anthropic spent this year in open conflict with the Pentagon over usage limits: the company would not permit its models to be used for mass surveillance of Americans or fully autonomous weapons, and Defense Secretary Pete Hegseth answered by designating Anthropic a national-security supply chain risk. On August 27, U.S. District Judge Rita F. Lin ruled that designation unlawful, finding the administration retaliated against the company's speech and violated its due process rights.

Four days later, the Pentagon shipped ChatGPT and Grok to the whole department, and Claude is still not aboard. Related litigation in D.C. is ongoing, per NPR, and nothing in last week's ruling forces the department to buy anything from Anthropic. Winning in court, it turns out, gets your blacklisting vacated; it does not get you a seat on the portal.

Why builders should care

First, the pattern. The largest employer in the United States just standardized on a multi-model portal: one accredited environment, several interchangeable frontier models behind it, picked per task. A War Department official framed the goal as "access to a diverse suite of AI capabilities from across the resilient American technology stack," which is procurement-speak for refusing vendor lock-in. If you sell AI into any regulated industry, this is the reference architecture your buyers will point at.

Second, the distribution math. OpenAI and xAI just gained a captive channel to 3 million professionals who will use these tools daily, at work, for free. Habits formed on GenAI.mil walk out the door into the civilian economy, the same way a generation of office workers learned Excel because their employer said so. Model choice increasingly happens at the institution level, not the individual level, and institutions are choosing in blocks of millions.

Third, the precedent on guardrails. Anthropic drew a line on surveillance and autonomous weapons, lost the contract, won the court case, and still is not on the platform. Every frontier lab watched that sequence. Whatever a vendor's published usage policy says, the practical question is now what happens to the ones who enforce theirs against their biggest customer.

Key Takeaways

• The Pentagon added OpenAI's ChatGPT Mil and Starshield AI's Grok for Government to GenAI.mil on August 31, joining Google's Gemini, the portal's only model since December 2025.
• The portal has more than 1.7 million users, and ChatGPT Mil is built to scale to the department's full workforce of over 3 million.
• Everything is accredited at Impact Level 5: Controlled Unclassified Information only, in a separate environment from consumer products, after DISA and NSA security review.
• ChatGPT Mil serves GPT-5.4 Terra today with GPT-5.6 Terra coming, per DefenseScoop, and targets document-heavy planning, policy, and logistics work.
• Claude remains absent four days after a federal judge ruled the Pentagon's supply-chain-risk blacklisting of Anthropic unlawful; the court win vacated the label but bought no contract.
• The multi-model, single-environment portal is becoming the reference architecture for institutional AI purchases, made in blocks of millions of seats.

Sources: Department of War release, DefenseScoop, Defense One, TechCrunch, The Hill, NPR