The AI platform at the Pentagon increased its user base from 80,000 to 1.5 million in just six months.

      TL;DR: GenAI.mil has reached 1.5 million daily users, up from 80,000 at its launch, following the Pentagon's implementation of Google Gemini and clarification of AI usage guidelines.

      According to the Pentagon’s chief technology officer, the generative AI platform GenAI.mil now attracts 1.5 million daily users within the Department of Defense, which represents nearly half of its total workforce of 3.5 million. Just six months ago, the platform had fewer than 100,000 users. Emil Michael, the undersecretary of defense for research and engineering, shared this information at a Hudson Institute event last week. At its launch in December 2025, only 80,000 personnel utilized the platform, with initial low adoption attributed to confusion regarding accessibility, functionality, and rules.

      Michael explained, “It wasn’t really clear where to go for it, what you could use it for, and the rules were unclear, so we just blew through that.” The breakthrough came with the introduction of Google’s Gemini on the Pentagon’s unclassified networks, leading to a surge in daily usage. Subsequently, OpenAI’s ChatGPT and xAI’s Grok were added, providing all 3.5 million DoD employees with access to several commercial AI models via a single portal.

      Tasks performed by personnel using AI mainly involve administrative duties. Michael noted that employees are engaged in drafting job descriptions, summarizing meeting notes, and creating budgets. A notable use case is congressional reporting, where tasks that previously took 200 hours can now be completed in just five hours by inputting source documents into the AI for report drafting.

      “There are more and more people realizing, ‘Oh my God, I could write a job description,’” Michael remarked, indicating a wide range of uses, from simple to complex tasks. The rapid adoption rate is striking, but not necessarily groundbreaking, as the Pentagon reports similar productivity increases seen in the corporate sector since the rise of large language models. Michael acknowledged that the goal is to catch up with what is now standard in the commercial sphere.

      The platform has progressed beyond basic chatbot functions. In April, it was announced that DoD personnel had created over 100,000 semi-autonomous AI agents using Gemini’s Agent Designer tool in under five weeks. These agents operate at Impact Level 5, the highest classification for unclassified sensitive information, managing tasks such as drafting after-action reports and analyzing operational data.

      The financial investment reflects the scale of adoption, with the fiscal 2027 defense budget proposing $54.6 billion for the Defence Autonomous Warfare Group, a significant rise from the previous year’s AI and autonomy budget of $13.4 billion. The Pentagon aims to integrate AI into every function, from administrative tasks to battlefield decision-making.

      However, this ambition has led to some tension. More than 580 Google employees, including senior DeepMind researchers, signed a letter in April requesting CEO Sundar Pichai to decline classified military AI work for the Pentagon, citing concerns about monitoring AI usage on air-gapped networks. This letter followed Google’s prior deployment of Gemini to the Pentagon’s unclassified personnel and ongoing negotiations for classified access under "all lawful uses" conditions.

      The Pentagon has since established classified AI agreements with seven companies, such as Google, Microsoft, Nvidia, and OpenAI, under conditions that intentionally supersede the safety protocols that Anthropic required before being excluded from Pentagon supply lines. The message is clear: the military sets the parameters for usage, not the vendors.

      The unclassified use of GenAI.mil, however, presents a different scenario. Tasks like writing job descriptions and preparing congressional reports involve low risks associated with AI’s limitations, such as hallucinations and factual inaccuracies. An error in a draft report can typically be identified by a human reviewer. The Pentagon has not provided information on error rates, accuracy levels, or any internal quality assessments of GenAI.mil’s outputs.

      Michael characterized the platform’s growth as organic. The Pentagon shared case studies illustrating how employees utilized GenAI.mil, disseminating this information throughout the department. Familiarity with consumer AI tools outside of work also contributed to personnel’s understanding of how chatbots function when they accessed GenAI.mil.

      Currently, five out of six military branches have designated GenAI.mil as their primary enterprise AI platform. The increase from 80,000 to 1.5 million users in just six months marks the fastest enterprise AI deployment in government history. Whether the quality of the output justifies the rapid adoption remains an unanswered question for the Pentagon.