ウクライナがドローンと地上ロボットのみでロシア陣地を占領

President Zelenskyy announces a historic first: a Russian position taken entirely by unmanned systems. A CSIS report details how AI is already changing Ukraine's battlefield and where the limits remain.

For the first time in the war in Ukraine, the Ukrainian military has captured an enemy position using only ground robots and drones, without a single infantry soldier involved. President Volodymyr Zelenskyy announced the milestone on Monday during an address to workers in Ukraine's defense industry.

The operation was carried out without infantry and without casualties on the Ukrainian side, with Russian occupiers surrendering, Zelenskyy said. The ground robotic systems involved included the Ratel, TerMIT, Ardal, Rys, Zmiy, Protector, and Volia. In the first three months of 2026, these systems completed more than 22,000 missions along the front lines.

But "unmanned" doesn't mean "autonomous." A report from the Center for Strategic and International Studies (CSIS) notes that during a comparable operation near Lyptsi, north of Kharkiv, in December 2024, the ground robots were still manually remote-controlled. Full autonomy - systems that independently find, select, and engage targets - doesn't exist on the Ukrainian battlefield, according to CSIS.

AI is already making a measurable difference in key areas

The AI applications currently deployed cover specific functions: reconnaissance, target recognition, and navigation. None of them handle the full kill chain from target identification to strike. But the impact is already measurable.

The clearest example is autonomous terminal guidance, known in Ukrainian military jargon as "last-mile navigation." According to the CSIS report, which draws on dozens of interviews with Ukrainian military personnel and defense manufacturers, autonomous navigation boosts FPV drone hit rates from 10 to 20 percent up to 70 to 80 percent. The reason is straightforward: the system doesn't need a stable radio link or a highly skilled pilot, both of which are frequently unavailable on the battlefield.

Instead of eight or nine drones per target, one or two are often enough, CSIS found. The company The Fourth Law offers a navigation module for $50 to $100 that's already in mass production. The most widely used system, VGI-9, can also track moving targets at speeds up to 80 km/h.

Target recognition now reaches up to two kilometers

Automatic target recognition (ATR) has improved from a range of 300 meters to an average of one kilometer in combat and up to two kilometers under optimal conditions, according to CSIS. AI models can identify tanks, artillery, vehicles, and infantry, and are also designed to spot decoys and camouflage that fool human eyes.

The Ukrainian system ZIR consists of a hardware module roughly the size of a bar of soap that can be mounted on different drone platforms. The software was trained on publicly available data. In 2024, the Ukrainian armed forces procured 10,000 AI-enhanced drones for the first time. With total drone production at around two million units that year, that's a small share, but it marks the beginning of systematic AI procurement.

Acoustic detection and text analysis are underrated AI tools

Beyond image analysis, Ukraine is deploying AI in less visible ways. The acoustic detection system Zvook identifies drones by their sound at distances up to 4.8 kilometers, delivering results within 12 seconds to the central situational awareness platform Delta. Each station costs around $500, according to CSIS, with a false alarm rate of just 1.6 percent. The system covers roughly 20,000 square kilometers.

The text analysis platform Griselda processes intercepted Russian communications and military group chats. According to founder Oleksiy Teplukhin, the AI replaces 99 percent of human work in transcription and semantic analysis. These figures come from the manufacturers themselves, however, and haven't been independently verified.

Small models, cheap chips, and fast training keep costs down

Ukrainian defense companies are deliberately training small, specialized AI models on limited datasets rather than building large general-purpose systems, CSIS reports. These models run on inexpensive chips installed directly on the drones and can be updated quickly. Software encryption is considered a strategic advantage: while adversaries can copy hardware within weeks, encrypted AI software significantly slows down reverse engineering.

Training soldiers on autonomous systems takes between 30 minutes and a day, according to CSIS. Tasks that once required extensive flight hours can now be taught to a much wider pool of operators.

Swarms and true autonomy remain out of reach

The CSIS report identifies two major hurdles for the next stage of development: extending autonomous capabilities to ground, sea, and underwater systems, and achieving drone swarms where multiple drones communicate and coordinate independently. Both are still in the early experimental stage.

Humans remain in charge of attack decisions, the report states. Ukrainian forces follow a human-in-the-loop approach where operators can override autonomous functions at any time. The AI models themselves, however, are still black boxes, which makes their use in life-or-death situations risky. Full autonomy, according to the report, remains a distant goal that runs up against technological, legal, and ethical limits.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.

Subscribe now