Fully autonomous drones appear to have killed human soldiers on the battlefield for the first time, according to a senior figure in Ukraine’s defence industry. If accurate, the incident would mark a significant shift in the use of artificial intelligence in war.
The account centers on a test carried out about two years ago with 10 AI-controlled quadcopter drones, described as “Terminator” drones. Alexander Kokhanovskyy, a drone maker who said he supplied the technology, told New Scientist that the system was used in a front-line experiment during the war in Ukraine. He said Russian soldiers were killed during the test.
Kokhanovskyy said the drones were launched toward the front line and flew several kilometres before switching into an autonomous attack mode. In that phase, he said, an AI model searched for targets and engaged them without any human oversight.
“We tried it,” he said at a press event hosted by the Ukrainian embassy. “It’s a test. We never implemented it [more widely].”
According to his description, the drones had no live data link to operators once the autonomous mode began. He said the vehicles flew on their own, could not be monitored through video during the attack, and were left to strike whatever they detected in the area.
After the test, human-piloted drones were reportedly sent in to inspect the outcome because operators had no direct record of what the autonomous aircraft had hit. Kokhanovskyy said the follow-up inspection found “a couple of soldiers” and “one truck,” and that this was taken as evidence the automated drones had killed the soldiers.
He said the test was conducted by an unnamed military unit near Bakhmut and Chasiv Yar during a Ukrainian counteroffensive push. Kokhanovskyy was not present for the operation, and the Ukrainian Ministry of Defence did not respond to questions about the reported test or the legal status of fully autonomous weapons.
The claims come as militaries increasingly use AI to process battlefield data, identify possible targets and automate parts of weapons systems. In most current systems, however, a human still makes the final decision to attack.
Ukraine’s own rules reportedly ban AI from making the final interception decision, even though automated tools are used earlier in the targeting chain. Kokhanovskyy said officials are discussing whether those restrictions should be relaxed as the technology develops.
The report also lands amid wider international concern about lethal autonomous weapons. The United Nations Secretary-General has called for a ban on systems that can kill without human involvement, while UN officials have warned that such weapons could raise serious legal and ethical problems by removing human judgement from combat.
Researchers and policy experts quoted in the report said the development would be deeply troubling if confirmed. Mariarosaria Taddeo of the University of Oxford argued that letting machines decide to kill strips responsibility from warfare and should be prohibited. Anthony King of the University of Exeter said fully autonomous attacks are technically possible, but added that systems retaining humans in the loop may still be more effective militarily.
Kokhanovskyy said the autonomous project has not advanced further because of Ukraine’s current rules. He now leads Aero Center, which is developing autonomous interceptor drones designed to stop incoming Russian Shahed drones before they reach cities and infrastructure. Even there, he said, current Ukrainian policy still requires humans to verify targets at the final stage.
The reported battlefield test is not the first claim that autonomous drones may have killed people in war, but it is presented as the most direct account yet of AI alone causing fatal casualties in combat.