The war that erupted in February 2022 has become a laboratory for autonomous drones. What began as a conflict fought with tanks, artillery and missiles now sees swarms of unmanned aerial systems that can locate, track and strike targets with little or no human input. Both Kyiv and Moscow have turned to private innovators, foreign partnerships and rapid prototyping to field ever more self-reliant machines.
In the early months, Russia struggled to produce large quantities of one-way attack drones and looked to Tehran for expertise. Over time it built a domestic pipeline capable of delivering loitering munitions such as the V2U. Ukrainian engineers, meanwhile, leveraged a fledgling drone industry and private defense firms to create a parallel ecosystem. By early 2025, Russian analysts recovered a V2U prototype equipped with a compact Nvidia processor that ran AI-driven decision software, lacking any external radio link—an unmistakable sign of full autonomy. Later that year, newer variants arrived without any communication antennas, confirming that the launch-to-impact chain could be completed without human oversight.
Field tests that blur the line between remote control and machine decision
November 2025 delivered a vivid illustration of hybrid autonomy when a Ukrainian first-person-view drone lost its telemetry link yet still managed to strike a likely Russian tank. An operator had selected the general area, but the onboard AI guided the final approach, proving that AI-assisted guidance can compensate for disrupted communications. A similar pattern emerged in July, when Russian forces used an AI-enabled drone to hit a gas station in Zaporizhzhia, killing three civilians. Operators chose the fuel depot as a target, but the drone’s internal software determined the exact impact point, showcasing a limited but consequential form of autonomous targeting.
Russia has also been integrating seeker-type packages into its Geran series—clones of Iran’s delta-wing Shaheed-136—allowing the computer to take over during the terminal phase after a human selects the strike zone. Ukraine, on the other hand, has focused on modular AI chips that can be bolted onto a wide range of platforms. One such module steers a drone for the last 500 meters to a recognized target, and the manufacturer claims more sophisticated versions will soon enable complete autonomy. Former defence minister Mykhailo Fedorov announced in July that Kyiv had fielded a drone capable of selecting and engaging targets in Russian-occupied Crimea without any radio connection, although mass production remains limited.
Reports from two years ago also mention fully autonomous quad-copter swarms used near Bakhmut, which reportedly eliminated Russian troops. While the drones were pre-programmed with waypoints, the question of whether they independently chose individual soldiers remains unsettled. The most recent visual evidence of autonomous attacks appeared on 19 September 2026, when a Russian drone set fire to a building in Dnipro, and on 20 September 2026, when a similar system ignited the Gazprom Neft refinery in Moscow, underscoring that fully self-guided weapons are already causing strategic damage.
Data as the engine of the AI arms race
Beyond the hardware, the conflict generates a torrent of battlefield imagery that is reshaping AI development. Ukrainian command structures process between 12 and 15 terabytes of raw drone footage each day, creating one of the most extensive real-combat datasets ever assembled. Each successful strike, each failed mission thwarted by electronic warfare or weather, feeds back into training cycles, allowing algorithms to learn vehicle shapes, camouflage patterns and movement habits in a way laboratory data never could.
Companies such as Perennial Autonomy have turned this bounty into operational tools, exemplified by the Hornet system that fuses computer-vision and rapid target recognition to hunt Russian logistical convoys. Similar AI-enhanced software is being mounted on inexpensive thermal cameras, turning ordinary off-the-shelf hardware into smart sights capable of tracking and engaging threats. The technology is spilling over into air defence, where autonomous interceptors now identify and shoot down Shahed-type attack drones without human direction.
Ukraine is also extending AI autonomy to other domains: underwater vehicles like Sea Trident and Marichka, ground robots under the Brave1 umbrella, and maritime drones that can navigate independently when communications are jammed. This diversification reflects a broader strategic insight—when electronic warfare can cut the link between pilot and platform, self-reliant navigation and targeting become decisive assets.
Ethical fault lines and the looming threat of swarms
Both sides acknowledge the tactical edge that autonomy offers, yet their legal frameworks differ sharply. Ukrainian defence policy mandates that a human must verify a target before the final interception, preserving a “human-in-the-loop” stance. Russian doctrine, by contrast, stresses the ability to intervene or deactivate systems but does not require a person to confirm each strike, effectively allowing human-on-the-loop operations.
The next frontier, however, may render these distinctions moot: swarming drones that communicate with one another, alter their flight paths on the fly and collectively decide which target to engage. A swarm of a few to several thousand autonomous units could overwhelm existing command-and-control structures, making it extremely difficult to enforce proportionality or protect civilians.
As the war continues to generate data, refine algorithms and proliferate smarter weapons, the international community faces an urgent dilemma: how to codify responsibility, enforce humanitarian law and decide at which stage a machine may be allowed to pull the trigger. The battlefield of 2026 already shows that the race for AI-driven drones is as much a moral contest as it is a technological one.
