The Pentagon sees generative AI as a force multiplier for an overstretched workforce, while security officials and some insiders see a rollout that must prove it can be trusted before it becomes routine.

The campaign began with GenAI.mil’s launch in December, initially offering a specialized version of Google’s Gemini. The department has since enrolled more than 1.7 million of its roughly 3 million military and civilian personnel, a sign of rapid institutional uptake even as the platform’s early rollout caught some employees off guard.¹

That December launch was so abrupt that a pop-up urging workers to use AI was initially mistaken by at least one Army office for a possible cyber intrusion, according to reporting on the rollout. Defense Secretary Pete Hegseth nevertheless set an uncompromising expectation: “AI should be in your battle rhythm every single day.”²

Now the department has added OpenAI’s ChatGPT Mil and Starshield AI’s Grok for Government to GenAI.mil. Both are approved to handle Controlled Unclassified Information at Impact Level 5, and Pentagon officials say the tools can take on document-heavy work in planning, logistics, administration and contracting. ChatGPT Mil, the department said, brings “a familiar commercial experience into the Department’s secure environment, tailored to warfighter needs.”²

OpenAI frames the change less as a battlefield shortcut than a way to reclaim staff time. Its government chief, Joe Larson, said the service would make unclassified work more efficient and free personnel to focus on the mission.² The Pentagon’s case for adding Grok is broader: multiple frontier models, it argues, reduce dependence on any one vendor and create a more resilient AI ecosystem.³

But the rush carries a warning label. The Defense Counterintelligence and Security Agency cautioned in June that unauthorized or embedded “shadow AI” can create “a massive, unmonitored attack surface” as new risks outpace safeguards and governance.³ The Pentagon’s bet is that tightly controlled, in-house access can beat that danger—before users turn to less secure alternatives.