UN negotiators have until the end of this year to agree a legally binding instrument on autonomous weapons, built around the principle that humans must stay in control of lethal decisions. In southern Ukraine, human operators use that control to hunt civilians.
On 4 August, a Russian drone hunted a street trader through a market in Kherson. Yuriy, 52, had begun laying out crates of tomatoes, cucumbers and aubergines beside his van when he heard the familiar buzzing overhead. His wife ran, and the drone turned towards him instead. He held up a handful of garlic to its camera, trying to show the operator that he was a vegetable seller, not a combatant. It detonated anyway, leaving him with concussion and shrapnel wounds. Yuriy gave his account to the BBC from a hospital bed.
Footage this clear is rare. Andrii Kovanyi, a spokesman for the Kherson police, told the Kyiv Independent that attacks of this kind have been a daily occurrence in the city for two years, and that few are captured in such detail. The operators fly from the occupied bank of the Dnipro into the part of Kherson that Ukraine holds, and Russian units publish their own drone footage of these attacks as a matter of routine. Ukrainian officials say the attacks double as training for operators before they are sent east, and the Prosecutor General has described the footage being recirculated in Russian propaganda on those terms.
Nothing in that sequence was autonomous. A person watched a live feed, understood what he was looking at, and initiated a lethal explosion anyway.
For fifteen years, the legal debate about drones and military artificial intelligence has been organised around the reverse of that scenario, the prospect of a machine that selects its own target and leaves nobody to prosecute. That prospect gave rise to the concept of the “responsibility gap” and continues to drive the campaign for a treaty on lethal autonomous weapon systems. Yet the civilian killing documented in Kherson has been carried out not by autonomous decisions but by fully attributable human ones, and attribution is not what is missing. The difficulty now is finding a court that can lay its hands on a defendant.
Where the law stands
In 2023, the UN Secretary-General and the President of the ICRC jointly called on states to conclude a legally binding instrument on autonomous weapons by 2026. António Guterres has for years described the delegation of life and death decisions to machines as “morally repugnant”.
That year has now arrived, and the Group of Governmental Experts operating under the Convention on Certain Conventional Weapons has one session left, from 31 August to 4 September, before it reports to the Convention’s Seventh Review Conference in November. Its mandate is to formulate, by consensus, the elements of an instrument. Consensus is precisely the difficulty, since a single state can halt the process, and Russia has repeatedly blocked it.
The General Assembly, which does not require consensus, has been more decisive. Resolution 79/62 passed in December 2024 by 166 votes to three, and a further resolution drew 156 states in late 2025, asking the CCW to complete its elements with a view to future negotiation. A resolution is not a treaty, and the negotiation it anticipates has not been convened. The UK holds that existing humanitarian law suffices, a position restated around last year’s Strategic Defence Review, which observed that “The UK’s competitors are unlikely to adhere to common ethical standards” without saying what the UK proposes to do about it.
What Kherson tests
The Kherson campaign offers the sharpest available measure of what human involvement in targeting is worth. In May 2025, the UN’s Independent International Commission of Inquiry on Ukraine concluded that Russian drone attacks on civilians along the right bank of the Dnipro amounted to the crime against humanity of murder, and probably to forcible transfer of population as well. The attacks were, in the Commission’s words, “widespread, systematic and conducted as part of a coordinated state policy”. Nearly 150 civilians were killed in the ten months examined, and ambulances were among the vehicles struck. Further reports in September and October 2025 extended the same findings across more than 300 kilometres of the right bank, into Dnipropetrovsk, Mykolaiv and Kherson. A senior hospital official in the city likened the operators’ conduct to playing a video game.
Every one of those strikes involved continuous human control, which the treaty campaign seeks to preserve. It supplied the intent for the crime rather than preventing it.
The case for a treaty
Supporters of a treaty have a ready answer, and it is a good one. Kherson demonstrates that human beings commit atrocities, which is exactly why we should not hand the same discretion to machines that cannot be prosecuted at all. That reply is correct on its own terms, and nothing here should be mistaken for the claim that meaningful human control is a bad idea. The campaign asks a single principle to carry more protective work than it can bear, and negotiators have concentrated on it partly because it is the portion of the problem on which states could plausibly agree. An instrument is still worth having. It simply would not have reached the market in Kherson, because nothing there was delegated to a machine.
Support for that reading comes from the other end of the spectrum. Ramón Reichert, writing in Dialogues on Digital Society last year, notes that what humanitarian organisations call a responsibility gap is understood inside military organisations as a tolerance spectrum, meaning an acceptable error rate rather than a legal defect. Contested reporting on the Israeli targeting systems known as Lavender and Gospel points the same way, describing machine-generated target lists passed to human reviewers at a tempo that makes review largely nominal. Kherson alone would carry the point. Read alongside it, the two suggest that keeping a person somewhere in the chain restrains very little by itself.
The responsibility gap has become an enforcement gap
Early discussion of autonomous weapons assumed that the central difficulty would be evidential, that as targeting decisions migrated into software, we would lose the ability to reconstruct who did what to whom. The war in Ukraine has produced the opposite result. Drone warfare has turned out to be the most thoroughly self-documenting form of killing yet devised, and the Commission of Inquiry built much of its case from more than 300 videos and over 600 Telegram posts, a great deal of it material the perpetrators uploaded themselves in order to terrify the people still living there.
Nor is there any shortage of applicable law. Article 51 of Additional Protocol I prohibits attacks directed against the civilian population and indiscriminate attacks. Article 8(2)(b)(i) of the Rome Statute makes intentionally directing attacks against civilians a war crime. Neither provision cares what the weapon was.
The deficit lies elsewhere, in the absence of a forum holding the people who hold the controllers. In June, Ukrainian prosecutors named ten servicemen of the 404th Motor Rifle Regiment over drone attacks on civilians in Kherson oblast and served notices of suspicion on them in absentia. Naming them is as far as the process can go. The International Criminal Court has jurisdiction over crimes committed on Ukrainian territory and no realistic prospect of a defendant in the dock. The one new institution European states have actually built for this war, the Special Tribunal for the Crime of Aggression against Ukraine, agreed with the Council of Europe in Strasbourg in June 2025, reaches senior political and military leadership and the decision to invade. It does not reach the operator watching Yuriy through a camera. That was a defensible choice, made to fill a jurisdictional hole the ICC could not, but it tells you where the accountability architecture of this war has been pointed, and it is not at the level where most of the civilian killing is now decided.
The same ranking applies further upstream. Sotoudehfar and colleagues argue in Defense & Security Analysis that the greatest present risk to civilians comes from cheap, poorly guided systems made at volume and passed to proxies. Export controls and minimum guidance standards would probably buy more civilian protection this decade than any agreed definition of autonomy.
Why this reaches Belfast
Belfast has more than a spectator’s stake in this, given that the Thales plant in the east of the city built the NLAW anti-tank weapons that shaped the first months of the Ukrainian defence and has since March 2025 been contracted to produce 5,000 air defence missiles for Ukraine under a package worth up to £1.6bn. A separate £176m order for lightweight multirole missiles was justified in part by the need to counter drones, placing east Belfast within the counter-drone economy that this war has created. That is reason enough to expect our MPs to hold a position on the November Review Conference, and if it produces another set of elements and no negotiating mandate, the loss will be smaller than it looks.
There is an older local connection. In December 2014, the NI Human Rights Festival brought Noel Sharkey to St Mary’s University College to lecture on drones and human rights, and Slugger carried the listing. A little under twelve years on, the capability he warned about has largely arrived. What Kherson shows is that the harm did not wait for it.
What follows from all this concerns transitional justice, and that is where it touches Northern Ireland. Accountability for killing is one of the conditions on which any settlement afterwards becomes liveable, because a society cannot come to terms with what it cannot attribute. Northern Ireland has spent three decades demonstrating that deferred accountability does not quietly decay. It compounds, and it resurfaces on someone else’s timetable. Kherson is generating an archive of crimes with identifiable victims and, in principle, identifiable perpetrators, in a war that will eventually end at a table where somebody will propose that the past be set aside in the interests of peace. What happens to that archive will shape the aftermath far more than the wording agreed in Geneva next month.
The real failure would be to keep treating the autonomy of the weapon as the whole of the moral problem, when the evidence is already on Telegram, and the courtrooms are the thing we have not built. Yuriy held up a handful of garlic to a camera to show that he was harmless, and a person at the other end of the feed watched him do it.
Sources: The Kherson attack is reported by BBC News, ‘Why did Russia strike a Ukrainian vegetable seller with a drone?’, 4 August 2026, https://www.bbc.co.uk/news/articles/cn4n03xg981o; and the Kyiv Independent, ‘‘It started flying right at me’ — Russian drone seen targeting vegetable seller at Kherson market in horrific footage’, 4 August 2026, https://kyivindependent.com/graphic-warning-horrific-video-shows-russian-drone-taunting-striking-vegetable-seller-at-kherson-market/. The Commission of Inquiry findings are in ‘They are hunting us’: systematic drone attacks targeting civilians in Kherson (A/HRC/59/CRP.2), 28 May 2025, extended by the Commission’s update of 22 September 2025 and its report to the General Assembly (A/80/497), 27 October 2025. Also relied on: UN General Assembly Resolution 79/62, 2 December 2024; Additional Protocol I, article 51; Rome Statute, article 8(2)(b)(i); Ramón Reichert, ‘Autonomous occupation’, Dialogues on Digital Society 1(3) (2025) 368; and Saba Sotoudehfar, Jeremy Julian Sarkin and Mohamed Zied Chaari, ‘Cheap drones, costly consequences’, Defense & Security Analysis 42(1) (2026) 139. Remaining sources are linked in the text.
El Cavador is a Slugger reader from Belfast with a particular interest in education
Discover more from Slugger O'Toole
Subscribe to get the latest posts sent to your email.
