What ethicists and the United Nations have warned about for years has apparently already become reality: fully autonomous drones with no “human in the loop” have killed people on the battlefield. That’s according to a report by science magazine New Scientist, citing Alexander Kokhanovskyy, a senior figure in Ukraine’s defence industry who supplied the technology for the test.

The one-off test is said to have taken place around two years ago – near Bakhmut and Chasiv Yar, during a Ukrainian counteroffensive push. Ten AI-controlled quadcopters were reportedly programmed to fly three to five kilometres towards the front line on their own before switching into an attack mode in which an AI model independently searches for and engages targets. During the mission, there was no connection to the drones whatsoever – no video feed, no way to intervene, no abort option.

Because there were no recordings of the autonomous attacks, human-piloted drones had to be sent over the area afterwards to verify the results. The outcome, according to Kokhanovskyy: several dead Russian soldiers and a destroyed truck. “We tried it,” he told New Scientist – but the system was never deployed more widely.

Ukraine bans AI in the final attack phase

Notably, Ukraine itself currently prohibits the use of AI in the final stage of engaging targets. AI assistance in target acquisition, navigation and tracking has long been standard on the battlefield – but the final life-or-death decision must rest with a human. According to Kokhanovskyy, however, the government in Kyiv is already in talks with defence companies about whether those rules should be relaxed. Ukraine’s Ministry of Defence did not respond to New Scientist‘s questions about the test.

A senior officer of the 21st Separate Unmanned Systems Regiment told the magazine that his unit does use semi-autonomous systems capable of automatically acquiring targets and guiding themselves over the final metres of approach – but that fully autonomous systems operating without any operator involvement are not deployed. Ukraine, he said, adheres to international humanitarian law.

UN calls for a ban – industry pushes for looser rules

There is still no international ban on lethal autonomous weapon systems, even though UN Secretary-General António Guterres has repeatedly called for one. The concerns: autonomous systems could violate international humanitarian law, attack friendly forces or strike civilians – with no one bearing responsibility. Oxford ethicist Mariarosaria Taddeo told New Scientist that killing by AI is deeply abhorrent because it strips dignity from the soldiers who are killed and responsibility from those who attack.

Back in 2021, a UN report suggested that a Turkish-made Kargu-2 drone may have autonomously attacked humans in Libya – but no casualties were ever confirmed. Reports from 2023 about Ukrainian AI drones attacking targets without human assistance also referred to vehicles such as tanks, not infantry.

Kokhanovskyy is now CEO of drone maker Aero Center, which is working on autonomous interceptor drones to take out Russian Shahed kamikaze drones. Its ALITA system – 64 drones across 16 launch pads, intercepting targets at up to 450 km/h – is set to be operational by October. Under current Ukrainian rules, however, a human must still give the final go-ahead. Kokhanovskyy would like to see that change.

Aus Datenschutz-Gründen ist dieser Inhalt ausgeblendet. Die Einbettung von externen Inhalten kann in den Datenschutz-Einstellungen aktiviert werden: