WASHINGTON, July 21 — The Pentagon has added ChatGPT Mil, a specialized military build of OpenAI’s chatbot, to its GenAI.mil portal, giving US troops and defense civilians another generative AI tool for controlled unclassified work. The deployment, announced by the Department of Defense on July 21, follows work that began in February 2026 to adapt ChatGPT for handling controlled unclassified information (CUI).
An OpenAI official said in June that the launch was expected in “early July.” ChatGPT Mil joins Google Gemini, the platform’s first tool when GenAI.mil went live on December 9, 2025, and Grok for Government. ChatGPT Mil is engineered for scale, built to support more than 3 million DoD personnel.
The intended workload is document-heavy and unclassified — planning, policy, logistics and administration — with the aim of accelerating routine tasks and freeing time for critical projects. A DoD official, speaking anonymously ahead of the announcement, said the portal’s value lies in choice: Gemini is preferred for search or scraping sources, while ChatGPT is better suited to text-based work. Under the hood, the military version is a separate engineering effort from the commercial model, with additional security functions.
CUI cannot be processed on commercial systems, so the environment had to be built from scratch — compute resources, user tracking with unique security conditions, data compilation and anticipated user requests. One former defense official involved in such tests described the process as “arduous,” likening it to passing security guardrails run by both DISA and the NSA: DISA builds the environment, NSA probes for vulnerabilities.
Neither OpenAI nor Pentagon officials intend to use these models for combat, they said in background talks. The technical reality is limiting: ChatGPT requires connectivity and large computing power, which makes it unsuitable for combat. That stance sits awkwardly with the DoD’s broader argument that it should be able to use AI products as it sees fit — a point of friction with model maker Anthropic.
The July 21 announcement also coincided with training at Joint Base Pearl Harbor-Hickam, Hawaii, where sailors from the U.S. Pacific Fleet were drilled on building AI agents by a U.S. Air Force captain from the Pentagon’s Chief Digital and Artificial Intelligence Office. The session was documented by Navy Mass Communication Specialist 2nd Class Christopher Sypert. What to watch next is the pace of model onboarding.
Preparing a frontier model for GenAI.mil is no quick job, and the Pentagon’s appetite for multiple vendors suggests more names will follow. For now, the detail that matters is access: troops get a familiar commercial interface, but one that has passed through the security guardrails first.
