After months of security testing, Pentagon adds an OpenAI product to the list of AI tools approved for unclassified work.
Updated: 5:29 p.m.
The Pentagon’s main portal for AI tools now offers ChatGPT and Grok for Government alongside Google Gemini.
A Defense Department official said the OpenAI product, dubbed ChatGPT Mil, would bring new strengths to GenAI.mil, where troops and defense civilians can use generative-AI tools to work on controlled, unclassified information, or CUI.
“Gemini, I like for search functionality or scraping certain sources. Chat for more, like, text-based work,” said the official, who requested anonymity to speak before the official announcement on Monday. “We want to have access to all these different models, so people, if they prefer one for a certain task, they can use it.”
Work began in February to create a version of ChatGPT that could be used for CUI. In June, an OpenAI official said they believed it would be deployed in “early July.”
A DOD press release slated for release today said ChatGPT Mil would bring “a familiar commercial experience into the Department’s secure environment, tailored to warfighter needs...ChatGPT Mil supports document-heavy unclassified work across the Department, including planning, policy, logistics, and administration, and is built for scale to support more than 3 million Department personnel. This deployment accelerates routine tasks and frees up time for more critical projects across the Joint Force.”
One OpenAI official, who requested anonymity to speak before the official announcement, offered ideas for how the military might use its special version of ChatGPT. Operators at Northern Command might use it to streamline planning, resource allocation, or disaster response with FEMA or other parts of the federal government. Or the Joint Staff’s logistics directorate might use it to simplify paperwork and documentation. The official also suggested that logisticians at combatant commands “would be a natural audience, with uses such as reconciling supply and maintenance updates, drafting transportation requirements.”
In background talks, OpenAI and Pentagon officials said they don’t want to use models on the GenAI.mil platform for combat. The Pentagon has argued that they should be able to use AI products as they like, a sticking point in disagreements between the department and AI model maker Anthropic. And OpenAI has said that ChatGPT’s need for connectivity and large amounts of computing power doesn’t make it a good fit for combat anyway. The department is continually testing and seeking a number of other tools for those types of operations and actions.
Customized for security
Like the GenAI.mil version of Gemini, the military version of ChatGPT has additional security functions built in.
“When you're dealing with CUI within the department, it's obvious this is stuff that cannot be on a commercial model. We have to build it in a separate sort of environment. So it's engineered in a way to be separate from the commercial model,” said the defense official.
Prepping a frontier model for use on GenAI.mil is an “arduous” process, said one former defense official who was involved in such tests. It requires building out the “environment”—that is, finding the compute resources, keeping track of the users and their unique security conditions, compiling the data that the model will use, and anticipating the sorts of requests that users may have.
“Think of having to pass security guardrails for both DISA [Defense Information Systems Agency] and the NSA,” the former defense official said.
As DISA builds the environment, NSA looks for ways that attackers might invade it—or the model might escape it, the official said.
The Pentagon launched GenAI.mil on Dec. 9 with one tool: Google’s Gemini. Around the same time, Pentagon officials signed an agreement to bring Elon Musk’s xAI tool to the platform, but no formal deployment has since been announced. On Monday, the Pentagon announced that Grok for Government, an AI product of Musk's Starshield AI, was now available on GenAI.mil.
Joe Larson, who leads OpenAI’s government sales, said, “ChatGPT on GenAI.mil will make unclassified work throughout the Department more efficient, freeing up time and resources so DoD personnel can focus on the mission. This integration is grounded in our national security principles, pairing access to powerful tools with clear safeguards, meaningful human judgment, and accountability for high-stakes decisions. We’re excited to see what the Department’s extraordinary workforce builds with it.”
Facts Only
* The Pentagon added ChatGPT Mil to the GenAI.mil portal.
* GenAI.mil is a platform for generative AI tools used for controlled, unclassified information (CUI).
* Google Gemini was launched on GenAI.mil on December 9.
* Grok for Government, a product of Starshield AI, is now available on GenAI.mil.
* Work on the military version of ChatGPT began in February.
* ChatGPT Mil is designed for use by more than 3 million Department of Defense personnel.
* The tool is intended for tasks including planning, policy, logistics, and administration.
* The military version of ChatGPT is engineered to be separate from the commercial model.
* Security guardrails for the environment are provided by DISA and the NSA.
* Both OpenAI and Pentagon officials stated the platform is not for combat use.
Executive Summary
The Department of Defense has expanded its GenAI.mil portal to include ChatGPT Mil and Grok for Government, joining the previously deployed Google Gemini. This ecosystem allows military personnel and defense civilians to apply generative AI to controlled, unclassified information (CUI) to streamline administrative, logistical, and planning tasks. By diversifying the available models, the Pentagon aims to provide users with a variety of strengths—such as Gemini's search capabilities versus ChatGPT's text-based strengths.
The deployment follows extensive security testing by DISA and the NSA to ensure the AI environment remains isolated from commercial models. While there is a clear push to accelerate routine bureaucracy for millions of personnel, there is an explicit agreement that these specific tools are not intended for combat operations. This boundary is driven by both technical constraints regarding computing power and connectivity, and a desire to maintain strict controls over AI's role in kinetic actions.
Full Take
The strongest version of this narrative is that the US military is prudently integrating commercial innovation into its bureaucracy to eliminate "paperwork" friction, thereby increasing the agility of the Joint Force without risking lives via autonomous combat systems. It presents a controlled, phased adoption of frontier models within a secure perimeter.
The narrative leans heavily on the "administrative efficiency" frame to sanitize the integration of powerful, third-party black-box models into the heart of national security infrastructure. By emphasizing the "unclassified" nature of the work, the focus is shifted away from the systemic risk of relying on private corporate entities (OpenAI, Google, xAI) for core government functions. There is an inherent tension between the Pentagon's desire for total control over AI usage and the operational realities of using proprietary models owned by commercial vendors.
Patterns detected: none
This echoes the historical pattern of "vendor lock-in" within the defense industrial complex, where the government becomes dependent on proprietary software that it cannot fully audit or replicate. The unstated assumption is that "unclassified" work remains sufficiently decoupled from "combat" work, despite the fact that logistics and planning are the foundational precursors to any kinetic action. The second-order consequence is the gradual outsourcing of cognitive labor—and the institutional knowledge associated with it—to a handful of Silicon Valley firms.
Bridge Questions:
1. If the military becomes dependent on these models for logistics and planning, does the distinction between "unclassified work" and "combat operations" remain meaningful?
2. What happens to operational continuity if a commercial provider's terms of service or corporate stability shifts?
Counterstrike Scan: A coordinated campaign would likely exaggerate the "AI-powered military" to project omnipotence to adversaries or incite fear of "robot wars" in the public. This content does not match that pattern; it is a measured announcement of administrative tool adoption.
Sentinel — Human
This article reads as well-researched reporting on a complex policy development, demonstrating an ability to weave technical updates with strategic implications, leaning towards human journalistic synthesis.