The U.S. Army has stood up a task force dedicated to building artificial intelligence agents that can "hunt" for threats inside Department of Defense networks, according to DefenseScoop, which reported on remarks from Army cyber chief Lt. Gen. Christopher Eubank.
The idea is straightforward even if the execution isn't: military networks are enormous, and the volume of activity flowing across them is far more than human analysts can inspect. AI agents — software that can take actions on its own rather than just answering questions — could in theory patrol those networks continuously, looking for intrusions.
What makes the effort notable is how cautious the Army sounds about its own project. Eubank described a daily process of deciding where the machines stop and people start. "Each and every day, we sit down as a group, we figure out the guardrails we're going to apply to these agents, [and] we ask ourselves: Is it risk that a human should be answering, or is it risk that an agent can answer?" he told DefenseScoop, adding that as of now, humans remain responsible.
A second concern is that the defense could become the vulnerability. DefenseScoop reported that an Army official warned about agents expanding the service's exposure rather than shrinking it: "We need help in implementing agents securely, so that it doesn't inadvertently increase our attack surface... If we have a bunch of agents roaming around, and they're vulnerable, that just made it worse because if the adversaries attack our agents, now we're exposed."
The Army is also looking for agents that are fast and that don't run up token costs — the per-use fees charged for the text AI models process, which can escalate quickly at military scale.
It matters because one of the world's most sensitive computer networks is about to be patrolled partly by software that can act on its own, and the Army is openly acknowledging it hasn't yet solved how to keep that software from becoming a target itself.