Autonomous AI agents are being adopted rapidly on the U.S. Defense Department's unclassified network.

TechRadar reported on May 4 local time that the department used an agent-designer version of Google Gemini on its GenAI.mil platform to build more than 103,000 semi-autonomous agents in less than 5 weeks. It is now adding more than 20,000 new tools each week.

These agents are run about 180,000 times a week. That is about 25,000 times a day on average. A session means a user runs an agent once.

The most-used agents handle repetitive tasks such as drafting after-action reports, writing official staff documents, analysing images and reviewing finance and strategy documents. Field personnel are creating tools directly within the network to automate routine digital work without separate, traditional programming knowledge.

Robert Malpass (로버트 말패스), deputy chief digital and AI officer for intelligence at the Defense Department, said at the INSA Spring Symposium that anyone across the department can build and use advanced AI tailored to their work context.

The system has received an operational approval at impact level 5 and runs on an unclassified network within a defined security and oversight scope. Still, concerns are also being raised about the speed at which automated tools are spreading. Some cite cases in which poorly controlled agents outside the Pentagon deleted systems or disrupted services, and operated without clear human approval.

Defense Department leadership sees it as difficult to slow the pace of technology adoption further. Andrew Maipas (앤드루 메이프스), acting chief deputy chief digital and AI officer at the Defense Department, said AI is accelerating the pace of technological change and that it should not take 5 to 10 years to bring new technology into the military.

The case shows a rapid spread of a trend in which internal military users build AI tools themselves and put them to work immediately. It also highlights a shift in how AI is adopted in defense, with large-scale use taking place within the scope of security approvals.