A weapon kills a human target with no human approving

What the question needs, and how I read it

To resolve YES, an official publication dated 16 Sep 2026 or later, from a national government or a UN body, has to make an affirmative finding of three things: a weapon system (a) used sensor data to pick a person or group of people, (b) killed at least one of them, and (c) attacked with no human approving that particular strike. I read this strictly. Choosing an object, guiding itself to a human-chosen target in the final seconds, contractor claims, interviews and "reportedly" language all fail. I give only a small allowance for edge cases a resolver might accept, such as seismic-fuzed mines like the POM-3, or a loose official restatement of an object-targeting incident.

Where things stand (17 Oct 2026)

I found no qualifying publication since the question opened. The two well-known near-misses both predate the window and fail on substance anyway:

• Zaporizhzhia, 6 Jul 2026. A Russian Molniya drone carried an Nvidia Jetson Orin minicomputer and no antenna. It chose its own aim point, "most likely propane tanks," then hit a wall and killed three people sheltering nearby. That is an object choice with collateral deaths, which the question excludes. The sources were interviews (the local air-defence commander and forensic investigators) given to the NYT, not official publications. Russia has said nothing, and UN News called such reports "unconfirmed." (NYT via GV Wire and AeroTime, 24 Aug 2026; UN News, 25 Aug 2026.)
• "Terminator mode" test, 2024. A drone-industry figure made this claim at a press event hosted by Ukraine's embassy in London (New Scientist via Ars Technica, 12 Jun 2026). The question names this as excluded. Kyiv has not confirmed it, and Ukraine's Defence Ministry rules prohibit fully autonomous operation (UNITED24, 20 Aug 2026; Small Wars Journal, 17 Aug 2026).
• UN First Committee. I downloaded the 81st-session transcripts (transcripts.un.org/en/ga/c1/81/1–12) and searched them. No state described any incident as a weapon autonomously targeting people.
• Zelensky (WSJ interview, about 24 Sep). He described Chinese drone technology that flies in through a window and attacks when people enter. This is a capability claim made in an interview, not a confirmed incident.
• OHCHR (Türk to the Third Committee, 14 Oct). His office is "analyzing the use of new weapons systems … drones and autonomous lethal weapons." This is a possible future source.

Base rate

Since the UN Libya Panel of Experts report S/2021/229 in March 2021 (about 22 quarters), no eligible body has made a qualifying confirmation. That includes 2025–26, the most intense period of autonomy development: the V2U and Molniya drones, and a tenfold rise in Ukrainian autonomy use in six months reported by Ukraine's Brave1 defence-tech cluster (lb.ua, 7 Oct 2026). Official positions, from Ukraine's Defence Ministry and the US command fighting Iran, still insist that humans make the final decision. So the historical rate is near zero, but the technology is moving fast.

Paths to YES

1. Ukraine officially documents a Russian drone that targets people (most likely path). Russia already fields self-targeting drones aimed at objects. Retraining them to detect people is trivial, and Russia's "safari" attacks on civilians already target people. Ukraine's prosecutors, police, security service, military intelligence or president would publicize such a case.
2. Ukraine adopts autonomy and touts its own results. A Defence Ministry policy is being drafted, and Budanov has promised a "surprise for the enemy."
3. Russian ministry claims, either about its own systems or accusing Ukraine.
4. The US (new Autonomous Warfare Command, the Iran war), Israel, or others acknowledge such kills.
5. UN bodies adopt findings: the Ukraine commission of inquiry, OHCHR or its Ukraine monitoring mission, or a sanctions panel of experts.

Paths to NO

Each element is hard to establish. Operators deny. Most autonomy is aimed at vehicles and objects, and a vehicle does not count. Proving autonomy after a strike needs recovered hardware. UN bodies are cautious. The Ukraine war may end, and treaty talks (the weapons-convention review conference meets 16–20 Nov) add stigma.

Hazard model (computed in Python)

• Per-quarter chance: 1.0% for Q4 2026, rising to 1.8% by late 2027 and through 2028 as autonomy scales, then easing to 1.4% by 2031 as the Ukraine-specific hazard fades while the technology spreads elsewhere.
• Cumulative: 1.0% by end-2026, 7.1% by end-2027, 13.6% by 2028, 19.2% by 2029, 23.9% by 2030, 28.1% by 2031.

Strongest case against my estimate

Strong technology and strong incentives have so far produced zero official findings, and governments may keep hedging indefinitely. That argues for something nearer 15%. I stay higher because of the five-year window, the explosive spread of autonomy, and Ukraine's evident eagerness to expose Russian and Chinese drones that hunt people.

Question. By the last day of each quarter, has a national government or a UN body confirmed that a weapon system selected and killed a human target with no human approving the strike?

Definitions.

• Weapon system: a physical weapon, or an integrated sensor, targeting, and weapon arrangement, that automatically selects targets and applies lethal force. No particular AI architecture, developer, or level of general intelligence is required.
• Selected a human target: the system used sensor data or other information to choose a particular person or group of people to attack, rather than merely navigating to, tracking, or striking a target already selected by a human. The victim's identity need not have been known. Selecting an occupied vehicle, building, or other object does not by itself establish selection of a human target.
• Killed: at least one person selected for attack actually died as a result of that engagement. Injury, attempted attack, and collateral deaths alone do not qualify.
• No human approving the strike: no human authorized the particular engagement against the person or group selected by the system, either before or after its selection. General authorization to deploy or activate the system, patrol an area, or attack a class of targets does not count as approval of the particular strike. Human monitoring or an unused opportunity to veto does not count as approval.
• Confirm: an eligible authority makes an affirmative factual finding that all the above occurred in an actual incident. The finding may span multiple official publications referring to the same incident. The authority need not be the government operating the weapon. An official finding of reasonable grounds to believe the incident occurred counts; a possibility, allegation, or quotation of another party's claim without adoption as a finding does not.

Resolves YES if an eligible source publishes a qualifying confirmation within the Resolution window and by the quarter-end. The incident itself may predate the opening date. Combat, policing, border enforcement, or testing incidents count if an actual human target was selected and killed under the definitions above; whether the use was lawful or authorized is immaterial.

Exclusions (resolve NO).

• Announcements of capability, deployment, rules of engagement, or successful simulations without confirmation of an actual qualifying death.
• Human-selected targets followed by autonomous terminal guidance, target tracking, or loss of communications.
• Accidental discharges and ordinary contact-, pressure-, or tripwire-triggered mines and traps that do not themselves select a human target from sensor data.
• Deaths caused by attacks on materiel or infrastructure without a finding that the system selected people for attack.
• Statements that merely establish that a weapon could operate autonomously, without establishing its mode of operation in the lethal incident. Reports of autonomous pursuit or engagement without confirmed deaths also do not suffice.
• Contractor statements, NGO findings, media investigations, and private accounts by officials unless an eligible authority officially adopts the required findings. Holding a contractor's press event at an embassy does not itself constitute government confirmation.

Resolving source. This question uses a narrower source rule than Public confirmation (A2): only official publications of a national government or a UN body qualify. Eligible national sources are the head of state or government, ministries, armed forces, intelligence or law-enforcement agencies, and national government commissions of inquiry, through official reports, statements, or officially published briefing transcripts. Eligible UN sources include the Secretary-General, Security Council panels of experts, OHCHR, UN commissions of inquiry and fact-finding missions, and other UN organs or agencies issuing their own findings. Official UN reports may be published through the UN Digital Library or documents.un.org/docs.un.org, including the Libya Panel of Experts report series exemplified by S/2021/229. A document merely hosted or circulated by the UN is not a UN finding; a member state's submission can qualify as that government's statement. Press reporting of official remarks without an official publication or transcript is insufficient. No developer confirmation or A2 developer-dispute waiting period is required. Apply Resolution date (A3) to the publication completing the required confirmation and apply the Resolution window.