Ukrainian investigators recovered an unencrypted American AI module from a Russian Molniya drone that killed three civilians near a Zaporizhzhia gas station on 6 July, and read the code that guided its targeting, the New York Times reported. No human operator made the final call.

Commercially sold AI boards have been turning up inside Russian strike weapons for more than a year, and autonomy is the direction both armies are moving in. What changed is the evidence rather than the technology. Until now, autonomy had to be inferred from what a wreck lacked—an antenna, a radio trace.

Tetiana Bubynets, a 19-year-old accounting student at Zaporizhzhia National University, was declared dead at a hospital, one of her professors said. Oleksii Svirin, 41, and Roman Karpii, 48, later died of their injuries. The drone had attempted to navigate around an apartment building toward the gas station's propane tanks but struck a wall and detonated near people sheltering there.

What investigators found in the wreckage

The wreckage contained an Nvidia Jetson Orin minicomputer, an off-the-shelf AI module that costs a few hundred dollars. The chip was not encrypted. Ukrainian officials said they could read the terrain imagery loaded for visual navigation and the object-detection code that defined what the drone was trained to target, NYT reported. The drone carried no radio antennas and was flying in a wave of approximately six, none of them transmitting. A missing antenna alone would not prove self-targeting—a drone flying a preprogrammed route needs no antenna either, but it would have no use for an Nvidia chip.

Human operators had programmed the drone to fly toward the gas station and set the general target area before launch. Once near the site, the onboard system identified and selected the specific object—most likely propane tanks it had been trained to recognize—on its own, drone experts and military officials said. Vadym Kushnikov, a drone expert at the Kharkiv Aviation Institute, described the drone as "a preprogrammed tool trained in virtual reality to track specific objects," NYT reported.

Where else the Nvidia module has turned up

Kateryna Bondar, a senior fellow at the Wadhwani AI Center at the Center for Strategic and International Studies in Washington, told NYT that no earlier Russian strike by a self-targeting drone has been documented killing civilians, which would make 6 July the first such case on record. She called the Nvidia module "the best proof" that Russia was experimenting with autonomous AI.

The Jetson Orin is not new to the battlefield. The module is made by the American chipmaker Nvidia. Ukraine's sanctions on Russia's AI drone supply networks, imposed in August 2025, hit Russian, Chinese, and Belarusian drone developers and component suppliers—not Nvidia.

Ukrainian intelligence first identified the module in the V2U autonomous strike drone in June 2025, mounted in a Chinese Leetop A203 minicomputer. It has since reported Jetson-based AI modules in Russia's Lancet and Klin drones—the Klin an older Jetson TX2—and, most recently, in the S-71 Monokhrom cruise missile.

In those cases, the finding was the hardware, and what Ukrainian intelligence inferred from it. This time, the code itself was legible.

Nvidia confirmed to NYT that the photographed board was a Jetson Orin. The company said the minicomputer is a consumer-grade product "sold to students, developers and start-ups for a wide range of beneficial applications," that it does not sell the computer in Russia, and that it is widely available through resellers.

Zaporizhzhia as testing ground

Russia began test flights of AI-guided Molniya drones in May, and had begun testing self-targeting systems in the V2U several months earlier, some of those for last-mile guidance only, Ukrainian air defense experts told NYT. Local air defense officials said Russia has been aiming at civilian objects—pumps and propane tanks at gas stations—along with military recruitment centers while trying out autonomous drones in Zaporizhzhia. Why those targets were chosen for the tests is not clear.

Gas stations have been under sustained attack independently of the AI trials. Russia has destroyed approximately 200 across Ukraine since spring, around 50 in June alone, by UNITED24 Media's count.

In Zaporizhzhia, gas stations now prop plastic sheeting at odd angles around propane tanks to confuse the drones' image recognition, and Ukraine has turned to shooting down drones that electronic warfare cannot jam.

An American AI computer has turned up inside Russia's newest cruise missile. Ukrainian specialists has found an Nvidia Jetson micro-computer in Russia's new S-71 "Monokhrom" cruise missile, the Defense Ministry's Main Intelligence Directorate (HUR) reports.

S-71M “Monokhrom is the first fully autonomous long-range strike system designed to engage not only stationary but also moving targets. The system is designed to independently decide whether to destroy a target based on threat priorities. A significant problem for such autonomous systems is determining exactly what they are attacking — a military truck or a bus carrying children.

The discovery is the latest evidence that Russia's newest weapons still run on Western technology. Ukraine has already documented hundreds of foreign components in Russian missiles and drones reaching Moscow despite sanctions.

The AI chip is the standout find

The use of the Nvidia Jetson may indicate the application of AI technologies in the missile. This capability would let the weapon process visual or sensor data onboard to identify targets or navigate, functions that depend on exactly the kind of compact AI computing it provides.

The chip sits among many foreign parts. In total, 35 electronic components used by Russia in the missile were identified, making Monokhrom another Russian weapon built substantially from foreign electronics.

The same parts show up across Russia's arsenal

HUR's analysis linked the missile to Russia's wider weapons program. The identified components included parts of a passive radar seeker that Russia has begun installing on Geran-2 drones to detect and strike Ukrainian air-defense systems and radars.

Other finds tied the missile to Russia's newest strike systems. They included a Chinese Honpho TS130C-01 camera from the jet-powered Geran-4 drone and components of the active homing seeker of the Kh-47M2 Kinzhal aeroballistic missile.

The finding sharpens the sanctions case

HUR framed the discovery as an argument for tighter enforcement.

"The discovery of an Nvidia Jetson in a new Russian missile once again demonstrates the need to strengthen sanctions pressure and coordinate the efforts of the civilized world," HUR said.

HUR specialists and Ukrainian research institutions continue to study Russian weapons and trace the origin of their electronic components to strengthen measures to block them.