Ukraine’s one-time test of fully autonomous drones crossed a military AI threshold
Two years ago, near the front line around Bakhmut in eastern Ukraine, a small swarm of 10 quadcopter drones flew toward Russian positions, severed all communication with their human operators, and activated a function their developers called “Terminator mode.” Onboard AI searched for, identified, and engaged targets without any human approving the strikes. The drones killed Russian soldiers. They did so without a single human pulling the trigger or even watching.
This was not a routine battlefield event. According to Alexander Kokhanovskyy, a senior figure in the Ukrainian defense industry who supplied the technology, it was a one-time test. The project was never scaled into broader deployment, and the company he now leads, Aero Center, was not involved in the test itself. He spoke about the incident at a press event hosted by the Ukrainian embassy in London, first reported by New Scientist on June 10 (New Scientist; Ars Technica).
Today the event is drawing intense scrutiny from military analysts, ethicists, and international lawyers. It represents the first confirmed combat use of a fully autonomous lethal weapon system, a “human-out-of-the-loop” capability that the United Nations has spent years trying to define binding rules for.
The autonomy spectrum
To understand why this matters, it helps to distinguish between the three levels of autonomy in weapons systems. The vast majority of drones used in Ukraine and other conflicts operate under human-in-the-loop control, where a person makes or approves every targeting decision. Many of Ukraine’s AI-assisted FPV drones operate at the next level, human-on-the-loop, where AI handles terminal guidance toward a target a human has already selected. The Bakhmut test was different: it was human-out-of-the-loop, meaning the AI selected the target and executed the strike entirely on its own.
Kokhanovskyy’s “Terminator mode” worked by flying the drones to a pre-programmed area over the Bakhmut front line. Once there, they cut all radio links with operators, making them immune to jamming and spoofing. The onboard AI then independently searched for targets, classified them as hostile, and engaged them. The entire operation took roughly 10 minutes over a distance of 3 to 5 kilometers (2 to 3 miles).
Why it was not scaled
Kokhanovskyy explicitly stated the project did not progress beyond this single test because of Ukrainian rules and regulations. This creates an important distinction between what Ukraine’s military establishment sanctions and what individual developers and units may have experimented with independently.
Ukraine’s official position on autonomous weapons is clear. Danylo Tsvok, head of the Defence Ministry’s AI Research Center A1, told Reuters on June 12 that “AI will form a new paradigm of warfare” but his center also stated in May 2026 that Ukraine is “not pursuing fully autonomous combat systems” and that “human control over final decisions must be retained” (Reuters). The apparent contradiction between this official policy and the reported test suggests the incident occurred outside formal approval channels.
A threshold crossed
Military AI experts have long warned that one of the most dangerous thresholds in modern warfare would be crossed when a fully autonomous system first made and executed a lethal decision without human oversight. The Bakhmut test appears to be that moment.
Under the United Nations Convention on Certain Conventional Weapons (CCW), signatory nations including Ukraine have held years of informal discussions about lethal autonomous weapons systems, but no binding treaty exists. The test raises questions under International Humanitarian Law about distinction (can an AI reliably distinguish combatants from civilians?), proportionality, and accountability (who is responsible when an autonomous system makes a targeting error?).
The legal “accountability gap” is the most difficult unresolved question. If a human commander orders an airstrike that kills civilians, that commander can be held responsible under the laws of war. If an AI makes the wrong targeting decision, there is no clear chain of accountability.
What comes next
Both Ukraine and Russia are racing toward greater autonomy on the battlefield. Russia has reportedly operationalized its own autonomous combat drones. Ukraine’s Defense Minister Mykhailo Fedorov, who has pushed for equipping 100% of frontline drones with machine vision and AI, was appointed as Defense Minister in January 2026. The New York Times reported in May that Fedorov’s goal is to raise Russian casualty rates from roughly 35,000 per month to over 50,000 through autonomous systems.
The Bakhmut test was a one-time experiment that did not enter operational use. But the fact that it happened at all means the threshold has been crossed. The question is no longer whether fully autonomous weapons will be used in combat, but how the international community will respond now that they have been.
Sources: New Scientist (June 10, 2026); Ars Technica (June 12, 2026); Reuters (June 12, 2026); Small Wars Journal (June 12, 2026); The National Interest (June 12, 2026)