The Pentagon sees generative AI as a force multiplier for everyday military work, while security officials and critics of the rushed rollout see a technology that must be tightly controlled before it becomes routine.

GenAI.mil began last December with a specialized version of Google’s Gemini, part of an “AI-first” push backed by War Secretary Pete Hegseth. But the rollout was abrupt enough that some personnel reportedly mistook a launch pop-up for a possible cyberattack. Hegseth’s instruction was unmistakable: AI should be part of employees’ “battle rhythm every single day.”¹

Now the Pentagon has added OpenAI’s ChatGPT Mil and Starshield AI’s Grok for Government to the platform, opening the tools to a workforce of more than 3 million military and civilian personnel. More than 1.7 million users have already been onboarded. The department says the models are approved for unclassified work and Controlled Unclassified Information at Impact Level 5, with ChatGPT designed to handle document-heavy tasks in planning, administration, logistics and policy.²

The official case is speed: Grok is being pitched for tasks from acquisition-market research to supply-chain management, with reasoning modes and reusable workspaces meant to sharpen collaboration. ChatGPT, meanwhile, is meant to bring a familiar interface into a secured environment “tailored to warfighter needs.”³

That promise comes with a counterweight. The Defense Counterintelligence and Security Agency has warned that unauthorized “shadow AI” can create data-leak risks and an “unmonitored attack surface” beyond existing safeguards.² The Pentagon’s answer is to keep users inside accredited systems rather than leave them to commercial tools.

The expansion also reflects a procurement strategy, not merely a productivity upgrade. By running multiple leading models, the department says it can avoid dependence on one provider. Yet Anthropic’s Claude is absent after its public clash with the Pentagon over restrictions on autonomous weapons and surveillance — a reminder that the military’s AI race is also becoming a contest over who sets the boundaries.¹