Pentagon adds Gemini 3.1 Pro to GenAI.mil as usage passes 1.3 million active users

The Pentagon’s enterprise AI platform is getting Google’s newest frontier model.

Defense One reported on April 27, 2026 that users of GenAI.mil now have access to Gemini 3.1 Pro after several weeks of preview use. The same report says GenAI.mil is available to up to 3 million defense users and is actively used by more than 1.3 million of them.

That makes this more than a model availability note. It is a scale signal for government AI adoption.

What changed

Gemini 3.1 Pro is now available to defense users through GenAI.mil, according to Defense One’s interview with Pentagon Chief Data Officer Gavin Kliger and Google Public Sector CEO Karen Dahut.

The platform launched in December 2025 and has been expanding quickly. Defense One reports that GenAI.mil reached 500,000 users within a week and 1 million users within a month, citing Dahut.

The agent story is just as important. Users have reportedly built more than 100,000 AI agents through GenAI.mil using Google Cloud’s Agent Designer. Those agents operate in Impact Level 5 environments for sensitive unclassified data, according to the report.

Why it matters

Government AI adoption usually moves slowly because procurement, accreditation, data handling, and security requirements are difficult. GenAI.mil appears to be moving unusually fast.

For Google, this strengthens Gemini’s position in government and regulated environments. For the broader market, it shows that agent builders are not limited to startups and software companies. Defense users are now using natural-language agent creation for internal workflows, documentation, data-heavy processes, and operational automation.

That does not mean the use cases are risk-free. Defense environments have higher stakes around hallucination, data exposure, model behavior, auditability, and misuse. But the usage numbers indicate that the Pentagon is treating AI as a platform layer, not a side experiment.

Tool impact

Gemini’s score should treat this as a government-distribution win and a trust signal for regulated deployments.

The impact is not about consumer Gemini chat. It is about Google proving that frontier models, agent tooling, and cloud accreditation can be packaged for a demanding public-sector customer at large scale.

For enterprise buyers outside government, the lesson is practical: ask vendors how their models behave inside controlled environments, how agents are permissioned, and whether usage can be audited at the workflow level.

What to watch

The next question is whether GenAI.mil remains primarily a productivity layer or becomes a mission-workflow layer.

If the platform keeps adding models from multiple frontier labs, it could become a live test of multi-model governance in a sensitive enterprise environment. If agent creation continues at this pace, the hard problem will shift from access to oversight: who approves agents, who monitors outputs, and who owns failures.