The Test That Changed the Conversation
Air Force test pilots have used a tactical artificial intelligence system to evade a live missile threat during a controlled flight test—a milestone that moves AI-assisted aerial combat from simulation into real-world demonstration. The test, conducted by pilots from the Air Force Test Pilot School in collaboration with the Defense Advanced Research Projects Agency, required the AI system to detect the missile launch, assess the threat geometry, and execute evasive maneuvering without direct pilot input.
The pilots were in the cockpit and retained override capability throughout the test, but the evasive maneuver sequence was executed by the AI system in the time window available between missile launch detection and intercept—a window that, depending on the geometry, can be measured in seconds. Human reaction time alone is insufficient to consistently execute optimal evasive maneuvers against modern missiles; the test was designed to determine whether AI systems can fill that gap.
The Speed-of-Combat Problem
Modern beyond-visual-range air combat has evolved to a tempo that strains human cognitive and physical capacity. Missiles travel at multiple times the speed of sound and can execute high-G terminal guidance maneuvers that a defending aircraft's pilot must predict and counter. The defensive geometry—determining the optimal evasion direction, appropriate use of chaff and flares, and maneuvering parameters that degrade the missile's guidance without exceeding the aircraft's structural limits—involves rapid computation under extreme stress.
Experienced fighter pilots develop intuitive responses through thousands of hours of training and simulator work, but those responses are based on pattern recognition against known missile types with known performance envelopes. As adversary missile technology evolves and as multi-threat scenarios become more common, the cognitive demands on human pilots continue to increase. AI systems that can absorb real-time sensor data and execute optimal defensive responses represent a potentially decisive capability advantage.
DARPA's Air Combat Evolution Program
The test is part of DARPA's Air Combat Evolution (ACE) program, which has been developing AI for aerial combat since 2019. The program first gained wide attention with the AlphaDogfight Trials in 2020, where an AI system defeated a human pilot in simulated visual-range combat. Since then, the program has progressively moved from simulation to hardware, testing AI systems in actual aircraft rather than in computer models.
The progression reflects both the maturation of the underlying AI and the regulatory and safety process required to conduct AI-controlled flight test operations. Each expansion of AI authority in flight test—from simulation, to hardware-in-the-loop testing, to actual flight with safety pilots, to operational demonstrations against real threats—requires extensive safety case development and risk assessment. The missile evasion test represents a particularly significant step because it involved a real threat rather than a simulated one.
What Was Demonstrated
The test demonstrated several specific capabilities. The AI system had to correctly identify and classify the incoming missile threat from sensor data—distinguishing it from other flight objects and characterizing its guidance mode and performance envelope. It then had to compute an evasive response that accounted for the aircraft's current energy state, the missile's predicted intercept geometry, and the constraints of the aircraft's flight envelope.
The evasive maneuver sequence included both kinematic maneuvering—altering the aircraft's trajectory to degrade the missile's geometric advantage—and the deployment of countermeasures, timed by the AI to maximize their effectiveness based on the missile's estimated guidance mode. The integration of both physical maneuvering and countermeasure deployment in a single AI-computed response is more sophisticated than either capability alone.
The Path to Operational Systems
Translating a successful test demonstration into an operational Air Force capability requires a development pipeline spanning years. Safety certification, integration with production aircraft systems, development of human-machine interface protocols, and doctrinal work determining when AI authority is appropriate are all substantial tasks. The Air Force's Collaborative Combat Aircraft program—developing autonomous wingman drones that would fly alongside crewed aircraft—provides one near-term pathway for this technology. The tactical AI systems being developed for missile evasion in crewed aircraft could be deployed first in uncrewed systems where the stakes of AI errors are lower and where the authority question is simpler.
The broader strategic implication of the test is clear: AI is moving from a support function in military aviation—mission planning, data analysis, logistics—to a direct role in the physical act of aerial combat. How that transition is managed, and what limits are set on autonomous AI authority in life-or-death situations, are questions the military is actively working through.
This article is based on reporting by Defense One. Read the original article.
