A classic SAM is human-operated, tied to a radar or IR sensor and to a fixed or vehicle-bound site. An AI-driven air-to-air UAS can launch itself, seek a target and engage — without an operator and without a radar site. That is not a faster missile, but a new class.
| Axis | Crewed SAM | Autonomous air-to-air UAS |
|---|---|---|
| Control | Human in the loop, fixed site | Self-launch, seek, engage |
| Sensing | Radar/IR, jammable | AI EO/IR vision, resists EW |
| Numbers | Few, expensive | Cheap, attritable, mass |
| Kill | Proximity/HE warhead | Also kinetic ram-kill |
Flares and DIRCM defeat an IR seeker; chaff and RWR/MWS address radar. A kinetic ram-kill with AI vision presents no seeker to blind and no radar lock to warn on — and a swarm overwhelms a suite sized for a handful of threats (see Chapter 7).
This is not a concept: Anduril's YFQ-44A "Fury" (a Collaborative Combat Aircraft) fired an AIM-120 AMRAAM in June 2026, with the Lattice software autonomously finding, tracking and computing the intercept — the human only gave the final order to fire. Exactly the "human-in-the-loop now, autonomous later" trajectory; the simple 1970s IR SA-7 illustrates the other end of the same line.
A cheap drone downing a fast, manoeuvring manned fighter is far harder than intercepting a slow Shahed. The low/slow end is operational; the mass-autonomous-versus-manned-fighter end is a this-decade trajectory. The urgency is a tempo argument.