Pentagon’s AI platform surges from 80,000 to 1.5 million users in 6 months

The Pentagon’s generative AI platform, GenAI.mil, has exploded from just 80,000 daily users to 1.5 million in only six months, according to the Department of Defense’s chief technology officer. That figure now represents nearly half of the DoD’s total workforce of 3.5 million personnel. When the platform launched in December 2025, only a tiny fraction of that number was actively using it.

Emil Michael, the undersecretary of defense for research and engineering, revealed the growth statistics during a Hudson Institute event last week. He explained that the initial low adoption stemmed from widespread confusion about where to access the tool, what tasks it could perform, and what the governing policies were. “It wasn’t really clear where to go for it, what you could use it for, the rules were unclear, so we just blew through that,” Michael said.

The inflection point came when the Pentagon deployed Google’s Gemini on its unclassified networks. That move triggered a surge in daily usage. Since then, the department has also integrated OpenAI’s ChatGPT and xAI’s Grok into the same portal, granting every one of the 3.5 million DoD employees access to multiple commercial AI models from a single interface.

Most of the tasks personnel are using AI for remain administrative. Michael cited examples like drafting job descriptions, summarizing meeting notes, and building budgets. But he highlighted one particularly impactful use case: congressional reporting. Work that previously demanded 200 hours of staffing time can now be completed in just five hours by feeding source documents into the AI and letting it generate the draft. “More and more people are like, ‘Oh my God, I could write a job description,’” Michael said. “I mean, very simple things to more exquisite things.”

The speed of adoption is what makes this rollout notable, not the novelty of the use cases. The Pentagon is essentially replicating the same productivity gains that corporations have been reporting since large language models entered the mainstream. Michael acknowledged this openly. “It’s just a matter of trying to catch up to, in this case, what’s basic in the commercial world,” he said.

The platform has evolved beyond basic chatbot interactions. In April, the Pentagon disclosed that DoD personnel had created more than 100,000 semi-autonomous AI agents using Gemini’s Agent Designer tool in fewer than five weeks. These agents operate at Impact Level 5, the highest classification for unclassified sensitive data, and handle tasks like drafting after-action reports, analyzing operational data, and reviewing images.

Spending has kept pace with adoption. The fiscal 2027 defence budget requests $54.6 billion for the Defence Autonomous Warfare Group, a sharp increase from the previous year’s allocation of $13.4 billion for AI and autonomy. The Pentagon’s broader strategy is to embed AI across every function, from back-office paperwork to battlefield decision-making.

That ambition has not been without controversy. More than 580 Google employees, including senior DeepMind researchers, signed a letter in April urging CEO Sundar Pichai to refuse classified military AI work for the Pentagon. Their argument was that on air-gapped networks, Google cannot monitor how its AI is used. The letter came after Google had already deployed Gemini to the Pentagon’s unclassified workforce and was negotiating classified access under “all lawful uses” terms.

The Pentagon has since signed classified AI agreements with seven companies, including Google, Microsoft, Nvidia, and OpenAI. These agreements deliberately replace the safety restrictions that Anthropic insisted on before being ejected from Pentagon supply chains. The message to AI vendors is unambiguous: the military defines the scope of use, not the company.

On the unclassified side, however, the risk profile is different. Writing job descriptions and drafting congressional reports are tasks where AI’s known limitations, such as hallucination, factual errors, and confidently wrong summaries, carry relatively low stakes. A misquoted statistic in a draft report can still be caught by a human reviewer. What the Pentagon has not disclosed are error rates, accuracy metrics, or any internal assessment of the quality of GenAI.mil’s outputs.

Michael described the growth as organic. The department published case studies showing how employees were using the platform and distributed them across the organization. Exposure to consumer AI tools outside of work also helped, he said, because personnel arrived at GenAI.mil already familiar with how chatbots work.

Five out of six military branches have now designated GenAI.mil as their primary enterprise AI platform. The trajectory from 80,000 to 1.5 million users in six months is the fastest enterprise AI rollout in government history. Whether the quality of the output justifies the speed of the adoption is a question the Pentagon has not yet answered publicly.