Pentagon says GenAI.mil now has 1.5 million users, up from 80,000 in six months

The Pentagon is no longer treating generative AI like a pilot project. Emil Michael, the Pentagon's chief technology officer and undersecretary of defense for research and engineering, says 1.5 million Department of Defense personnel and employees are now using GenAI.mil every day.

That headline number lands hard because it is a reversal from where DoD started. When GenAI.mil launched in December 2025, it had roughly 80,000 daily users out of 3.5 million department workers. Michael says the platform’s early adoption was held back by uncertainty about where to go, what it could be used for, and what the rules were, so DoD “just blew through that.”

Here is the practical story behind the stats. GenAI.mil is the Defense Department’s hub for artificial intelligence programs for US military personnel and workers, built to put tools into hands that traditionally consume memos, paperwork, and process-heavy tasks. According to Michael, the platform saw fewer than 100,000 users six months earlier, and since then usage has skyrocketed. The growth matters because it signals that the limiting factor was not raw interest. It was friction: lack of clarity, lack of sanctioned use cases, and unclear guidance.

Michael laid out the timing and the mechanism. After GenAI.mil was introduced in December 2025, DoD then launched Google’s Gemini on its unclassified networks. From that point, he says daily users increased dramatically. In his framing, the Pentagon’s directives and internal communications did not just authorize AI, they made it legible. DoD also did case studies on what people were using GenAI.mil for, then spread those patterns across the department, which helped use cases proliferate.

If you are a board member, operator, or investor watching this from the commercial side, the second-order lesson is how rapidly adoption accelerates once governance gets operational. In other words, rules and “approved use” cannot stay abstract. They have to become workflow. Michael specifically points to paperwork as a common use. He describes users like “Oh my God, I could write a job description,” ranging from simple drafting to more complex work. He gives a concrete example: loading papers into the system so it can draft a congressional report that would otherwise take 200 hours of staffing time and do it in five hours.

That is the efficiency promise DoD originally said it would deliver, and it helps explain why this now looks like more than a novelty. When you reduce cognitive load, you do not just save time, you change what people can afford to do. Michael also connects adoption to a behavior shift: employees being exposed to AI outside work. That external exposure then raises internal expectations, and once DoD provides a sanctioned on-ramp, employees will pull the lever.

The Pentagon’s bigger AI push is also part of the backdrop. Michael ties the GenAI.mil growth to directives about what AI can and should be used for, and the department’s encouragement to roll out GenAI.mil and other AI platforms across its workforce. The Pentagon has also been requesting billions of dollars for next-generation AI and computers in the fiscal 2027 budget, reflecting that GenAI adoption is being treated as infrastructure, not experimentation.

And then there is the warfighting question. DoD is exploring how AI can be used in and around combat, while officials have assured guardrails and insisted humans remain in the loop. Still, they acknowledge the speed of future warfare may require AI to help process data and make faster decisions. For executives, that is a governance tightrope: the same systems that speed up drafting and analysis in offices may also, eventually, speed up decision-support in high-stakes environments, so the guardrails have to mature as usage expands.

So what should leaders in adjacent industries take from this? The Pentagon just demonstrated that enterprise generative AI adoption can go from “unclear” to millions of users quickly when three things line up: distribution through a central hub (GenAI.mil), a clear portal to credible models (Google’s Gemini on unclassified networks), and operational policy that makes use cases real. The strategic stake is not whether AI is useful. It is whether your organization can turn usefulness into repeatable, governed workflows at scale before competitors do.