Pentagon Links ChatGPT Mil and Grok for Government to GenAI.mil and Ends Exclusivity of Gemini
The Department of Defense's AI portal now runs three frontier models in parallel, reaching 1.7 million users among a force of 3 million military, civilian, and contracted personnel.
On Monday, August 31, the U.S. Department of Defense connected OpenAI's ChatGPT Mil and Starshield's Grok for Government to GenAI.mil, an internal generative AI portal launched in December 2025 that initially featured only Google's Gemini for Government. According to the Pentagon, GenAI.mil now has 1.7 million users from a total force of 3 million military, civilian, and contracted personnel, and in this round, it adds two competing frontier models available side by side on the same panel.
ChatGPT Mil is credentialed for work with Controlled Unclassified Information (CUI) at Impact Level 5, the DoD's cloud standard for sensitive unclassified data. The initial configuration includes chat, files, projects, and custom GPTs, covering planning, policy, logistics, and administration. Grok for Government enters with equivalent scope on the same model selection screen.
The End of Single-Account at the Largest IT Buyer in the U.S.
The concentration on a single vendor is the officially cited problem. In announcing Grok, the Pentagon stated that the goal is to "eliminate vendor lock-in and promote a broader American AI ecosystem." This move executes the design laid out in July 2025, when the Chief Digital and AI Office (CDAO) distributed contracts of up to $200 million per vendor to Anthropic, Google, OpenAI, and xAI, with a declared focus on agentic AI in defense flows. Anthropic, banned from the Pentagon in March, is the most visible absence on the shelf, despite having signed the original contract.
Placing three models under the same portal changes the commercial relationship. A single buyer, the CDAO, controls distribution, authentication, and auditing across 3 million terminals. This reduces part of the traditional value of enterprise accounts, as the vendor no longer builds a direct relationship with the end user. xAI's route to the portal was swift, from the contract in July 2025 to go-live in August 2026, and passed through Starshield, the SpaceX division credentialed for sensitive government data.
For CIOs in regulated sectors still debating how many LLMs to keep in production, the benchmark has become explicit. The largest technology buyer in the U.S. has decided not to choose but rather to standardize the interface, which weakens the single-vendor thesis in advisory.
How the Model Propagates Outside the U.S.
This design becomes a reference for allied governments. In the UK, the Ministry of Defence has been studying a similar architecture for sensitive unclassified data for months, and the Gemini for Government, ChatGPT Gov, and Grok for Government offerings now have a ready case for sale in Whitehall. In Brussels, analysts are evaluating a similar standard for the GovTech AI of the European Commission, with a hard difference: the transparency obligations of the AI Act, effective since August 2, create additional friction with American suppliers and may prevent a direct copy.
The other country that feels the immediate effect is India. A significant portion of the integration of these platforms into U.S. military and federal flows passes through the delivery centers of Accenture Federal Services, Booz Allen, and Leidos, as well as the offshore operations of TCS, Infosys, and Wipro serving public sector accounts. Each credentialed model on GenAI.mil translates into demand for engineers with clearance at American bases and for platform engineers at hubs in Bangalore and Pune. The vendor plurality also drives up integration costs, as each new model requires interoperability testing that offsets the decreased risk of dependence with more billable consulting hours.
What This Round Leaves Open
The question that the upcoming quarters will answer is how the CDAO intends to measure the comparative performance of the three models and whether this measurement will feed back into purchasing distribution or turn into a political weapon in the next contract dispute. If the Pentagon publishes usage metrics, task success rates, and cost per query with granularity per vendor, it will be the first time that a U.S. federal buyer produces such public comparisons of LLMs in a sensitive environment, with immediate effects on the enterprise pitches of OpenAI, Google, and xAI in the private sector.