GenAI.mil has moved from fewer than 100,000 users to 1.5 million daily users in six months, giving the Defense Department one of the clearest tests yet of how fast generative AI can spread inside a large enterprise.
When Emil Michael, the Pentagon's chief technology officer and undersecretary of defense for research and engineering, spoke at the Hudson Institute last week, the number that mattered was not abstract. According to Business Insider, Michael said 1.5 million Defense Department workers now use GenAI.mil every day, up from 80,000 users when the platform was introduced in December 2025. For a workforce of roughly 3.5 million people, that is a serious adoption curve, and it is happening inside an institution usually treated as the hard case for enterprise software.
The platform began with Google's Gemini on unclassified Defense Department networks. ChatGPT was added through the Pentagon's secure AI platform earlier this year, and Grok has also been brought into the department's AI stack through xAI's government work. The exact vendor map has shifted as the Pentagon has pushed the major model companies toward defense use, but the operating idea is clear enough: put commercial AI models in front of personnel, give them clearer rules, and let the useful work show up in daily routines rather than in slide decks.
The agent numbers are the part founders should read twice. TechRadar, citing reporting from Breaking Defense, reported in May that Pentagon personnel had built more than 103,000 semi-autonomous agents in less than five weeks using a version of Gemini's Agent Designer on GenAI.mil. Those agents were running about 180,000 sessions a week, or roughly 25,000 daily uses, on unclassified networks with Impact Level 5 authorization. The jobs were not exotic demonstrations. They included drafting after-action reports, assembling staff estimates, analyzing imagery, and reviewing financial or strategy documents.
That is why the story is bigger than another government software deployment. The workers making these tools are not all engineers. They are people inside the bureaucracy trying to remove dull staff work from their day. Robert Malpass, the Pentagon's deputy chief digital and AI officer for intelligence, described the shift at the INSA Spring Symposium as giving people across the department the ability to build advanced AI into their own context. The phrase sounds broad, but the examples are narrow enough to matter: reports, reviews, summaries, and workflow steps that used to eat hours.
The old assumption was that government would be the last place this kind of AI behavior would spread. Slow procurement, security review, cultural caution, and unclear authority usually make adoption grind. Michael's explanation to Business Insider was more practical. He said the department gave clearer direction on where to go, what the tools could be used for, and then circulated case studies showing what personnel were actually doing with them. One example he gave was a congressional report that might have taken 200 hours of staff time being drafted in five hours after workers loaded the relevant papers into the system.
That is a distribution lesson as much as a product lesson. Many AI workflow startups still talk as if the main problem is convincing employees that the tools are powerful. GenAI.mil suggests the sharper problem is showing workers one job they already recognize, in an environment where access, rules, and permission have been handled for them. Once that happens, the jump from chatbot to agent is not theoretical. It is the budget analyst or logistics officer building a tool because the alternative is another night of manual review.
The vendor competition is still unsettled. Google had the first clear advantage because Gemini was available at launch and because the agent tooling that produced the 103,000-agent figure came through Gemini's infrastructure. OpenAI's ChatGPT gives the Pentagon another familiar model for drafting, summarization, and analysis. xAI's Grok has been tied to a Defense Department contract worth up to $200 million, as Politico and other outlets reported last year. These contracts are not large by the standards of frontier AI companies, but the usage is valuable. A model that becomes routine for federal workers earns a kind of placement that is hard to buy later.
Anthropic remains the unresolved name in the room. The Verge reported in February that Anthropic's negotiations with the Pentagon had become a fight over use restrictions, including mass surveillance and autonomous weapons, and over a contract worth up to $200 million. That dispute sits awkwardly beside the rapid uptake of GenAI.mil because it shows the two sides of the same federal AI push. The Pentagon wants speed and broad lawful use. Some model companies still want red lines written into the work. The adoption numbers do not settle that argument. They make it more urgent.
For founders building workflow AI, the safer conclusion is not that every enterprise can copy the Pentagon. Most companies do not have a single platform for 3.5 million workers, nor a leadership chain that can turn direction into daily behavior. But the ceiling on adoption looks higher than many people assumed. If a security-heavy bureaucracy can move from 80,000 users to 1.5 million daily users in six months, the blocker in other large organizations is less likely to be worker curiosity. It is the unfinished plumbing around access, permissions, examples, and trust.
The Pentagon has not finished the rollout. It still has millions of workers outside daily use, and the harder questions around oversight, sensitive data, and military applications are not solved by a usage chart. But the first half of GenAI.mil's growth is already enough to puncture a lazy idea about enterprise AI: that adoption will creep forward only after years of committee work. Sometimes people just need a sanctioned place to start, and a task they are tired of doing by hand.
Also read: NewCore launches with $66 million to give AI agents proper enterprise identities • Nvidia sells $20 billion in bonds as AI demand outpaces its own cash generation • Sarvam AI is India's newest AI unicorn after raising $234 million from HCLTech
