Fully Autonomous Killer Drones, update
Sending killer drones into a geofenced kill zone to hunt and destroy people and vehicles has already happened.
The science and technology magazine New Scientist reported last week that Ukraine has already tested fully autonomous drones on the battlefield, marking a dramatic escalation in the automation of warfare and the erosion of human control over lethal battlefield decisions.
The test, according to New Scientist, took place two years ago, with quadcopters that, once launched toward the front line, switched into an AI mode called “Terminator.” The drones then searched for and attacked targets without human supervision. If accurate, it would be the clearest public account yet of fully autonomous lethal drones killing human beings in combat.
In March I wrote that the gap between semi-autonomous weapons and fully autonomous ones is narrower than most policymakers have wanted to admit. The systems I examined then still required a human to choose the target, but they could already navigate on their own, continue under jamming, and complete the strike once a person had set the mission in motion.
The new claim matters because it appears to remove that last human step. According to Alexander Kokhanovskyy, a professional gamer and wealthy game entrepreneur now with a drone startup, the drones were launched toward the battlefield and then switched into an onboard AI mode that searched for and intercepted targets inside a defined area.
He told New Scientist there was no connection to the drones once that mode engaged: no video feed, and no way for a human to monitor or veto what they attacked.
If that account is accurate, the machine was given authority to decide what to kill inside a bounded area and then act on that decision alone.
Ukrainian firms already field drones that can maintain course under electronic warfare, match onboard imagery to a human-selected target, and execute the final attack phase without continuous control. Those capabilities rest on cheap single-board computers, Nvidia Jetson modules when needed, and relatively modest neural networks running on the edge.
Ukraine has also built an unusually rich wartime data environment. Its Ministry of Defense maintains a Universal Military Dataset with more than two million hours of drone footage and millions of labeled military objects. Systems use that data to identify and classify Russian equipment in live video streams and command-and-control tools.
Given that infrastructure, the move from semi-autonomous targeting to bounded autonomous attack is more a doctrinal decision than an engineering breakthrough. The hardware exists. The models exist. The operational need exists. What changes is how much discretion commanders are willing to delegate to software.
The hardest technical problem is target discrimination: separating a soldier from a medic or a military truck from a civilian vehicle. That is why geography matters. A depopulated stretch of frontline terrain is a very different environment from a city. The more a commander can assume that anyone moving in a particular area is hostile, the easier it becomes to justify machine-selected engagements there.
In my March piece, I noted that fully autonomous weapons would allow a force to define a geofenced zone in which drones could hunt for any person or vehicle matching programmed rules.
That appears to be what Kokhanovskyy described. The operator launches the drones into a defined area. The AI then takes over and treats whatever it recognizes inside that area as targetable. The human no longer decides whether this specific person or vehicle should be attacked. The human decides that anything the machine identifies inside this box may be attacked.
International Humanitarian Law requires distinction, proportionality, and the ability to halt or adapt attacks. It does not explicitly say a human must make every final targeting decision. U.S. doctrine talks about “appropriate levels of human judgment over the use of force,” a phrase elastic enough to cover systems in which human judgment is exercised at the level of mission design rather than individual engagement.
Ukraine has reasons both to resist and to push autonomy. Errors that kill civilians largely kill Ukrainians, which is a real constraint. At the same time, Russia has more mass, uses heavy jamming, and has shown less concern for civilian life. That asymmetry creates pressure on Ukraine to move quickly toward systems that can preserve some discipline while keeping pace technically. It is also an inevitable future of war.