A senior figure in the Ukrainian defense industry has confirmed, in an on-record interview, that a fully autonomous drone selected and killed Russian soldiers on the front line near Bakhmut and Chasiv Yar roughly two years ago, in what amounts to the first publicly acknowledged lethal strike by a weapon that made its own targeting decision without a human in the loop.
Alexander Kokhanovskyy, a Ukrainian drone maker, told New Scientist at a press event held at a Ukrainian embassy that around ten AI-controlled quadcopter drones were dispatched to the front as part of a counteroffensive. Each was programmed to fly to the line, loiter for roughly ten minutes over a three-to-five-kilometer perimeter, and then switch to what Kokhanovskyy called "terminator mode," a fully autonomous target-and-attack setting in which the drone used an unidentified AI model to seek out and engage human targets, with no human watching its video feed and no option to abort. The aftermath was confirmed by a separate human-piloted drone sent to scan the impact zone. Kokhanovskyy described the result as "a couple of [Russian] soldiers" and a truck destroyed, according to Futurism's write-up of the interview.
The disclosure has been kept quiet for about two years, and Kokhanovskyy framed it as a one-off test that was not repeated. He was not present at the test, which was carried out by an unnamed Ukrainian military unit. The Ukrainian Ministry of Defence declined to answer New Scientist's questions about the strike or about Ukraine's current legal position on fully autonomous weapons. Other Ukrainian defense-company representatives at the same press event said Kyiv currently bans AI at the final targeting and intercept stage, and that the government is in talks with industry over whether to relax that line.
That official silence is the story. For years, militaries and arms makers have pointed to a maintained distinction: a human stays in the loop, or at least on the loop, when a weapon pulls the trigger. The Israeli military's loitering munitions, "fire-and-forget" drones that circle a target area before striking, are sold and deployed under the same general claim of human-supervised engagement. The United States is reported to be developing "Goalkeeper" flying drones and "Whiplash" naval drones capable of finding and engaging targets on their own, and US Project Maven-style tools already select targets inside the kill chain, in theory with a human confirming the shot. The Kokhanovskyy admission is the first time a senior figure has publicly stated that, in at least this case, the human was not in the loop at all.
It is also not an isolated line of evidence. A 2021 UN Panel of Experts report concluded that a Turkish-made Kargu-2 quadcopter "may have autonomously attacked" humans during fighting in Libya in 2020. The panel gave no detail on the source of that claim or on whether anyone was hurt, and the Libya case has remained thinly documented. New Scientist reported in 2023 that Ukrainian AI attack drones may have been finding and striking targets without human assistance, but only against vehicles, with no human casualties confirmed at that time. The Kokhanovskyy disclosure, if his account of the test holds up, moves that line from "may have" and "against vehicles" to a named, on-record description of a lethal autonomous strike against people.
The international response has not caught up. UN Secretary-General António Guterres said in May 2025 that "there is no place for lethal autonomous weapon systems in our world," but no treaty or enforcement mechanism is in place. The wider documented pattern of AI-assisted targeting has only deepened in the meantime. Reporting this year has linked the US military's use of Anthropic's Claude model to target selection in Iran, and according to Futurism, that targeting pipeline is believed to have contributed to the strike on the Shajareh Tayyebeh school that killed 168 children, teachers, and staff, a claim that other reporting has disputed. Whether the trigger was pulled by a person or by a model does not change what landed on the building. It does change who is answerable for it, and on what legal record.
For now, the threshold has been crossed, documented, and disclosed by one man in one interview. No government has confirmed or denied it. No international body has opened an inquiry. The next tell is whether Ukraine, or any other party that has used or is developing fully autonomous weapons, comes forward with its own accounting, or whether this disclosure stays where it is: a single on-record voice describing a single test, in a war that has otherwise normalized the slow, quiet arrival of machines that pick their own targets.