After more than four years of war, Ukraine has become the leading power in military combat drones, in which AI plays an important role. Officially, the Ukrainian Government prohibits its use in the final phase of target interceptionbut that has not prevented the national industry from carrying out real combat trials with drones completely guided by AI, without human intervention. As revealed to New Scientist Alexander KokhanovskyyCEO of the Ukrainian drone manufacturer Aero Center, Fully autonomous UAVs killed Russian soldiers during a battlefield test two years ago.
In the test, which is unrelated to his current company, pre-programmed quadcopter drones flew to an area on the front line and activated a ‘Terminator mode’ powered by AI that searched for and attacked any target within the indicated area, Kokhanovskyy explained.
Apparently, there was no video signal or any other element to show what the ‘Terminator’ drones located and attacked. But Kokhanovskyy told New Scientist that human-piloted drones sent to test the consequences found ‘a couple’ of dead Russian soldiers, leading to conclusion that fully autonomous drones had killed them.
The interview with Kokhanovskyy took place at a press event organized by the Ukrainian embassy in London. There, a Ukrainian military commander explained to the media that his drone pilots They only use semi-autonomous systems in which there are always humans making the crucial control decisions. He spoke of Ukraine’s commitment to ‘international humanitarian law’ and stressed that the army always acts with ‘great care in decision-making to avoid civilian casualties’.
The one-time nature of the experiment fits not only with the official position of the Ukrainian Government, but with the reality of the risks it presents. Sending fully autonomous drones to attack everything they find in a given area, without the intervention of any human operator, requires careful planning and carries the risk of ‘friendly fire’ incidents or attacks against non-combatant civilians. It is also unclear to what extent these fully autonomous quadcopters were effective at selecting and engaging targets compared to human-piloted drones.
To what extent are Ukrainian drones autonomous?
Fully autonomous weapons with the ability to ‘accomplish objectives independently or with minimal supervision in complex and unpredictable environments’ are not yet a reality in the war in Ukraine, according to Kateryna Bondarformer advisor to the Ukrainian Government, in her report for the think tank Center for Strategic and International StudiesCSIS, based in Washington DC But Bondar stands out the growing number of drones that integrate certain autonomous capabilities for navigation and, sometimes, targetingalthough human operators maintain overall control.
Ukraine and Russia are using many FPV drones for reconnaissance and attacks against vehicles and soldiers. Typically, these drones are controlled by trained pilots who wear virtual reality goggles to see from the drone’s point of view while aiming at enemy targets. There are also larger quadcopter or multirotor ‘bomber’ drones, capable of carrying heavier loads, either for supply missions or to drop explosives on enemy targets on the front.
Longer-range attack drones, similar to fixed-wing aircraft, can incorporate more autonomous decision-making capabilities. During 2025, Russia launched tens of thousands of drones to attack Ukrainian cities and positions, including drones Shahid originally supplied by Iran and increasingly manufactured in Russia. Shaheds are typically pre-programmed to fly automatically toward their targets, with little autonomous decision-making ability. But some variants of Shahed, such as Geran-2are equipped with microcomputers Nvidia Jetson Orin smuggled in, which provide onboard video processing and autonomous decision-making capabilities, including autonomous target recognition and retargeting.
To defend against these Russian Shahed drone attacks, Ukraine has deployed a hybrid air defensewith advanced systems and simpler solutions, such as cheap nationally manufactured interceptor drones. Some interceptor drone systems are designed to fly autonomously to the interception point and lock onto the target, although a human operator is still required to make the initial target selection and give the attack orderalways maintaining the ability to cancel the attack, according to the Ukrainian media United24.
Meanwhile, Ukraine has acquired the technological and industrial capacity to launch more than 5,000 drone strikes against Russian targets at distances greater than 20 kilometers each month, according to the Ukrainian Ministry of Defense. These medium and long range attack drones rely heavily on autonomous navigation capabilities due to Russian electronic warfare systemswhich can interfere with communication with human operators, as well as interference from GPSwhich can divert UAVs guided by this system from their objective. That AI-powered navigation has raised the success rate of Ukrainian drone attacks from around 10 or 20 percent to 70 or 80 percent.Bondar states in his report to CSIS.
As a whole, the Ukrainian defense industry has focused on training small AI models with small data sets to operate with the limited computing power of small, cheap chipsBondar points out. That approach has created AI-powered software for crucial autonomous functions such as navigation and target recognitionwhich can be packaged on separate hardware modules to install it on small FPV drones, long-range attack drones or even in the armed turrets of unmanned ground robots. So while fully autonomous systems remain rare, we can expect more drones and robots to incorporate specific autonomous AI capabilities that complement human decision-making on the battlefield.