The Air Force Is Planning for AI-Powered Hackers
16th Air Force has a four-part defensive campaign plan designed to blunt autonomous, agentic AI attacks on military networks.
By The Defense Desk
The U.S. Air Force's cyber command has published a defensive campaign plan built around a specific threat: AI agents capable of autonomously attacking military networks at scale. Lt. Gen. Thomas Hensley, commander of 16th Air Force/Air Forces Cyber, unveiled the framework at the annual Department of the Air Force Information Technology and Cyberpower conference in Montgomery, Alabama, calling frontier AI models a direct and present concern. [1]
The plan organizes defensive cyber operations into four lines of effort. The first — what Hensley called the 'bread and butter' of the mission — focuses on hardening systems and blunting adversary attacks through persistent network monitoring by airmen in security operations centers and network operations centers, cyber protection teams responding to incidents, and active threat hunting using specialized equipment. [1]
[1] The architecture of priorities reflects a triage logic — not every node can be defended equally, so the most consequential ones are defended first.
The third line may be the most operationally charged: 'proactive cost imposition.' Hensley was direct about its intent. [1] The line deliberately blurs the boundary between defense and offense.
[1] The goal: verify every user and, even when an adversary breaches the perimeter, prevent them from getting to where they want to go or accessing the information they seek.
What makes the plan notable is less its architecture than its stated adversary. Defensive cyber doctrine has long accounted for human operators running sophisticated intrusion campaigns. Designing explicitly against autonomous AI agents — ones that can probe, pivot, and persist without a person in the loop — marks a shift in threat modeling that most public-sector frameworks have yet to name directly.
Hensley acknowledged how quickly the threat landscape has changed. 'Last year was anybody talking about frontier AI models? Was anybody really talking about this kind of capability? No. Was anybody talking about frontier AI models six months ago? No,' he said. [1] He then posed the next question: 'These are the first movers, right? So who are the fast followers? They're the folks that have the second mover advantage because they can glean off of the work that the first movers have done. What more powerful capabilities and tools are they developing to do more and faster?'
The defensive plan follows the development of an offensive cyber operations campaign plan last year, though Hensley declined to describe that framework publicly. [1] Taken together, the two documents suggest 16th Air Force is positioning itself to operate across the full spectrum of cyber conflict — not merely to absorb attacks but to shape the behavior of those launching them.
The speed of the threat is the real variable. Military institutions are not typically built for the cadence at which AI capabilities are being developed and deployed. A campaign plan that is relevant today may require significant revision within months — which may be precisely why Hensley chose to speak about it publicly now.
The plan acknowledges something most defense bureaucracies have been slow to say plainly: the attacker does not have to be human anymore. Building networks that can resist an AI agent is a different engineering problem than building ones that resist a person — and the Air Force, at least publicly, is treating it as such. [1]