A Russian drone picked its own target. A few-hundred-dollar Nvidia board was inside.

Tetiana Bubynets, 19, was sheltering against a gas station wall on the edge of Zaporizhzhia when the Molniya strike drone came in. The drone had been trained to find propane tanks, and its software chose the final aim point. It missed, detonated against the wall, and killed her and two men, aged 41 and 48. The algorithm picked the target. The wall was where the people were.

Forensic teams recovered the targeting computer from the wreckage and identified it as a Nvidia Jetson Orin, a consumer-grade AI module Nvidia sells to students, developers, and startups for a few hundred dollars. The module carried no encryption, according to the New York Times reconstruction, and exposed both the terrain imagery the drone used to navigate and the code specifying which object classes it had been trained to attack.

Kateryna Bondar, a senior fellow at the Center for Strategic and International Studies, told the Times that the strike is the first documented case of civilian deaths caused by a Russian drone using fully autonomous targeting. A human operator launched the Molniya toward the gas station; the final aim point was selected onboard, by software, against a fixed target list.

Col. Serhiy Minaiev, the commander of Zaporizhzhia's air defense, told the Times: "Machines are making decisions to strike." The strike drone flew in a group of roughly half a dozen aircraft, none of which emitted radio traffic, per Ukrainian air defense officials.

Russian forces have used fiber-optic and jamming-resistant FPV drones for two years, and those platforms still rely on a human steering them to a building, a vehicle, or a foxhole. The Molniya's onboard model was trained to recognize a category: propane tanks, vehicles, antennae. It picked one when a human had put it in the area. Russia is understood to have begun AI-guided test flights of the mass-produced, fixed-wing Molniya in May 2026, according to the Times reporting, after months of self-targeting trials with an earlier V2U model.

Nvidia confirmed to the Times that the photographed modules were Jetson Orin minicomputers selling for a few hundred dollars. The company's statement described them as "consumer-grade products sold to students, developers, and startups for a wide range of beneficial applications" and "not designed for military purposes." Nvidia also acknowledged that pre-owned Jetsons are available through reseller channels and that it cannot track post-sale use. The company has said Jetson Orin modules are "not available in Russia" and, in the same statement, conceded that the modules can be reached through third parties. Ukraine's GUR intelligence directorate has separately documented the proliferation channels that pull consumer Western electronics into Russian strike platforms.

A few-hundred-dollar dev board designed for a robotics hobbyist is not, by itself, a weapon. Once it carries a targeting model, it is the brain of one, and the original seller can plausibly say it never knew. The Jetson could be swapped for any other general-purpose AI board at the same price point, and the question that lands is what end-use verification should apply when a consumer AI dev kit can be re-tasked, in a garage, into the targeting brain of a mass-produced autonomous strike drone, and what guardrails, export controls, or chain-of-custody rules the chip industry has not yet built.

The CSIS framing, the recovered hardware, and the unencrypted targeting code together make the strike at the Zaporizhzhia gas station a documented precedent. The Molniya entered mass production. The next AI-guided test flight is not hypothetical.