The Pentagon has added ChatGPT Mil to GenAI.mil, its portal for artificial-intelligence tools used by troops and defense civilians handling controlled, unclassified information. The OpenAI product joins Google Gemini and Grok for Government on the platform, expanding the department’s approved options for unclassified generative-AI work, Defense One reported.
A Defense Department official said ChatGPT Mil is intended to support document-heavy tasks across the department, including planning, policy, logistics and administration. A planned department announcement described the system as bringing “a familiar commercial experience into the Department’s secure environment” and said it was built to support more than 3 million department personnel.
The deployment follows months of preparation. Work on a version of ChatGPT suitable for controlled, unclassified information began in February. An OpenAI official said in June that deployment was expected in early July, before the product appeared on the Pentagon’s main AI portal.
The system is designed to operate separately from OpenAI’s commercial model. The Defense Department official said that controlled, unclassified information cannot be placed on a commercial model and that the military version therefore requires a distinct environment with additional security functions.
That preparation includes more than making the model available to users. A former defense official involved in testing similar systems said the process requires computing resources, user-account controls tied to individual security conditions, relevant data and planning for the requests personnel may submit. The Defense Information Systems Agency builds the environment while the National Security Agency tests how attackers might enter it or how the model might escape its intended boundaries, the former official said.
The security work reflects the requirements for placing a frontier model on a Defense Department system that handles controlled, unclassified information. The former official described the preparation as an arduous process involving the operating environment, user tracking, data and anticipated requests. The official said the system must pass security guardrails associated with both the Defense Information Systems Agency and the National Security Agency.
The initial uses described for ChatGPT Mil center on administrative and planning work rather than combat. Northern Command operators could use it to streamline planning, resource allocation or disaster-response coordination with the Federal Emergency Management Agency and other federal entities. The Joint Staff’s logistics directorate could use it to simplify paperwork, while combatant-command logisticians could reconcile supply and maintenance updates or draft transportation requirements, an OpenAI official said.
Pentagon and OpenAI officials have said they do not intend to use models on GenAI.mil for combat. OpenAI has also said ChatGPT’s reliance on connectivity and substantial computing power makes it a poor fit for combat operations. The department is continuing to test and seek other tools for those missions.
GenAI.mil launched on Dec. 9 with Google’s Gemini as its only tool. The Pentagon later announced that Grok for Government, an AI product from Elon Musk’s Starshield AI, was available on the platform. ChatGPT Mil’s addition gives department personnel another model for unclassified work and adds OpenAI’s system to the department’s growing collection of generative-AI tools.