AI is changing Russia’s drone war on Ukrainian cities, deputy air defense commander says

The most consequential detail in this report is not the jet engine. It is the machine vision. A one-way attack drone that can recognize and track a target on its own changes the economics of both offense and defense, and it does so in a way that no amount of additional jamming power can fully answer. For three years the dominant counter-UAS assumption in Western procurement has been that the electromagnetic spectrum is the kill chain's soft underbelly: jam the command link, spoof the navigation, and the drone falls out of the sky or wanders off. That assumption is now being retired in public, by a Ukrainian deputy air defense commander, no less.

The mechanics matter. Earlier generations of the Geran family, like the Shahed-136 lineage they descend from, were essentially cruise missiles with a crude autopilot: a pre-programmed route, a terminal dive, and a warhead. Their accuracy against anything other than a fixed coordinate was limited, and their vulnerability to electronic attack was high because everything depended on satellite navigation and, in some variants, a radio link back to a controller. Adding a jet engine raises speed and complicates interception by piston-engine aircraft and slow gun trucks, but it is the machine-vision terminal seeker that changes the engagement problem. A camera coupled to onboard processing can compare what it sees against a stored target signature or a terrain reference, correct its own course in the last seconds, and strike a moving or relocated target that jamming cannot move. The human operator, if there is one, becomes a supervisor rather than a pilot. That is the difference between a guided munition and an autonomous one, and it is the difference that air-defense planners have been dreading.

On the demand side, this accelerates a reordering of the counter-UAS market that was already underway. Radio-frequency jamming and spoofing remain cheap, scalable, and effective against the majority of drones in the air today, and no serious program is abandoning them. But the growth margin is shifting toward effectors that do not depend on the target emitting: radar and acoustic detection nets, electro-optical and infrared tracking with increasingly capable onboard processing, and hard-kill interceptors ranging from gun-based systems and guided micro-munitions to directed-energy prototypes. The sensor fusion problem, not the shooter problem, is where the engineering difficulty now sits. A defender must detect a small, fast, low-flying object, classify it, and commit an interceptor in seconds, often in urban clutter where collateral risk constrains the choice of effector. Machine vision on the offense raises the required performance of machine vision on the defense, and it raises the premium on automation in the command-and-control layer, because human reaction times are no longer the binding constraint.