The claim comes from one industry source at an embassy press event, with no video, and a serving commander says his unit uses only semi autonomous drones. That gap matters because there is still no internationally agreed definition of a lethal autonomous weapon.
A Ukrainian drone-industry CEO says his company once tested fully autonomous quadcopters that found and killed Russian soldiers without a human in the targeting loop. A serving Ukrainian commander quoted in the same reporting describes a different reality: his drone unit uses only semi-autonomous systems, with a human making every targeting decision.
The gap between the two accounts sits at the heart of a story that has been framed in some quarters as the moment autonomous weapons crossed a threshold on the battlefield. The threshold depends on a category, "lethal autonomous weapon system," that has no agreed international definition. Until someone defines it, the public cannot tell whether the alleged test was a genuine first, an ordinary semi-autonomous engagement mislabeled, or something that did not happen as described.
The claim comes from Alexander Kokhanovskyy, the CEO of Aero Center, who told New Scientist at a Ukrainian embassy press event in London that roughly two years ago an unnamed Ukrainian unit ran a one-time test near Bakhmut and Chasiv Yar. According to Kokhanovskyy, ten quadcopters were preprogrammed to fly three to five kilometers to a front-line sector, then switch on an AI targeting mode he calls "Terminator mode," which searched for and engaged targets with no human in the loop, no communications, and no video feed. Kokhanovskyy was not present at the test. A human-piloted aftermath check reportedly found "a couple of soldiers, one truck" dead. There is no recording of the autonomous engagement, and Aero Center did not exist at the time.
The Ukrainian Ministry of Defence did not respond to New Scientist's questions. Major Danylo Polozhukhno of the 21st Separate Unmanned Systems Regiment, part of the 3rd Army Corps, said he was unaware of any such test and told the outlet his unit uses semi-autonomous systems with humans always in the loop. Polozhukhno framed Ukraine's drone posture around international humanitarian law and care to prevent civilian casualties. That direct contradiction, between an industry CEO at a press conference and a serving commander in the same article, is the story.
"Terminator mode" is the CEO's marketing name for the AI targeting function, not a Ukrainian military program designation. The terminology matters because the test, if it happened as described, would push into a category of weapons the United Nations has tried, and failed, to define for more than a decade. UN Secretary-General António Guterres has called for a ban on lethal autonomous weapon systems. The U.S. Department of Defense defines them as weapons that, once activated, can select and engage targets without further human intervention. Most allies read against that definition when crafting their own policies. The category remains a moving target in international law.
The wider evidence on Ukraine's drone posture does not support a claim that fully autonomous systems are operating at scale. A March 2025 CSIS analysis by Kateryna Bondar, a former advisor to the Ukrainian government, concludes that full battlefield autonomy "is not yet present" in Ukraine. The deployed AI is partial: navigation, target recognition, last-mile approach, and ISR analysis. Ukraine's defence industry trains small models on small datasets for cheap onboard chips, and AI-assisted navigation has reportedly lifted long-range strike success rates from 10–20% to 70–80%, per the same report.
The Ukrainian military's monthly drone-strike count now runs above 5,000 at ranges beyond 20 kilometers, according to the country's Ministry of Defence. Many of those strikes depend on autonomous navigation because Russian electronic warfare and GPS jamming degrade human-controlled links. That is partial autonomy in the navigation layer, not the targeting layer. The difference is the part that decides whom to kill.
Kokhanovskyy told New Scientist the project has not progressed beyond the single test, and the practical limits he acknowledged were the reason: preprogrammed area attacks carry real friendly-fire and civilian-harm risk. Embassy press conference sources told the outlet that Ukraine currently bans AI at the final engagement stage, but that the government is in talks with defence companies about whether to loosen the rules. The policy landscape is moving, and a one-off experiment in 2024 should not be read as Ukraine's doctrine.
The episode is also a study in how a thin story travels. A single industry source at an embassy press event, with no video, no independent witnesses, and a direct contradiction from a serving military commander, has been picked up as a "first" in global coverage. The next test is whether the Ukrainian government, or any independent body, confirms or denies the account. Until then, the threshold the world was told was crossed was crossed only in a press conference, against a definition no one has agreed on.