UN Killer Robot Negotiations Collapse in Geneva: 128 States Settle for Two More Years of Study
Posted on 22nd Nov 2026 11:04:56 in Artificial Intelligence, Machine Learning
Tagged as: autonomous weapons, killer robots, UN Geneva, AI governance, defence AI, lethal autonomous weapons
Five days of negotiations in Geneva ended at dawn on Saturday without the one thing around 100 governments had come for: a mandate to negotiate a legally binding treaty on autonomous weapons. Instead, the 128 states party to the UN Convention on Certain Conventional Weapons (CCW) agreed to spend the next two years "elaborating" an instrument whose legal status remains undecided.
The Seventh Review Conference had opened on Monday with the International Committee of the Red Cross calling for binding rules within a year, a demand backed by UN Secretary-General Antonio Guterres and by Pope Leo. The final session ran past 4am local time after an all-night push, and still produced only a procedural compromise. Russia led the objections, delegates said, while the United States, Israel and several European defence-technology powers resisted anything that would bind them to a new protocol.
The result is the latest data point in a 13-year argument about whether machines should be allowed to decide who dies in war — and the clearest evidence yet that the consensus-bound machinery of the CCW cannot answer that question on its own.
What the Geneva Outcome Actually Commits States To
The mandate adopted on Saturday does not authorise negotiations on a treaty. It asks high contracting parties to "elaborate by consensus" on an international instrument, and it leaves the legal character of that instrument to be decided later. In practical terms:
- A group of governmental experts is tasked with developing the instrument and recommending its legal nature, including whether it should become a binding protocol to the convention.
- Nothing in the text guarantees that the eventual instrument carries legal obligations. Some of the yet-to-be-adopted provisions could remain optional.
- The work runs for two years, pushing any decision to 2028 — well beyond the end-of-2026 deadline Guterres had sought.
- The consensus rule stays in force, so any single state can still block adoption at the end of the process.
UN High Representative for Disarmament Affairs Izumi Nakamitsu described the outcome as a "significant achievement" while conceding its limits. "The path to or possibility to have a legally binding instrument is still open; it's not closed," she told reporters. The conference's Swiss chair called the outcome a significant achievement for humanitarian disarmament, saying it showed that "even in today's deeply divided world, multilateral diplomacy can still deliver concrete results."
The Red Cross did not share that reading. The ICRC said it "regrets" that the review conference "failed to agree to a clear and immediate negotiation mandate for a legally binding instrument to regulate autonomous weapons systems," adding: "While an opportunity has been missed, the imperative to act remains." The organisation wants unpredictable autonomous weapons, and systems designed to target humans directly, prohibited outright, with tight restrictions on the rest.
Who Wanted a Treaty, and Who Stalled It
Roughly 100 states arrived in Geneva backing binding rules. A bloc of seven — Austria, Belgium, Brazil, Ireland, Mexico, New Zealand and Panama — led the push for a treaty negotiated within one year. European and NATO states including Ukraine, along with China and Iran, expressed support for or willingness to consider a new protocol.
Brazil's delegate put the case bluntly on the floor: "These are not weapons of the future being discussed here, these are already in use and being further developed. We are late in developing these norms and rules to regulate these lethal autonomous weapons systems."
On the other side sat the largest arms producers. The United States proposed a five-year window to discuss nonbinding, optional guidelines rather than a treaty. Israel opposed binding rules. South Korea, Japan, Turkey, the UK, France and Germany — all exploring AI-enabled weapons — resisted being bound by new law and argued instead for voluntary measures. Some delegations argued that existing international humanitarian law already covers the problem.
Diplomats inside the room said Russia deliberately filibustered, objecting to the meeting's agenda on day one and then to rules of procedure in ways that forced the cancellation of formal sessions. Other participants were described as seeking a genuine consensus statement; Russia, by several accounts, was not working in good faith. The exclusion of several non-member countries from the negotiations also triggered a rare protest in the chamber, with Egypt's delegation shouting at the chair on the final day.
Why One Country Can Block 127 Others
The CCW operates by consensus. All 128 high contracting parties must agree before the convention adopts anything, a threshold so high that the forum has not passed a single new law regulating weapons in 30 years. When states wanted to ban cluster munitions, they gave up on Geneva and negotiated the 2008 Oslo Convention outside the UN framework. Campaigners argue autonomous weapons may now need the same treatment.
"There was no political will to really regulate these weapons," one European diplomat said. "No will commensurate to the urgency and need."
Max Tegmark, chair of the Future of Life Institute, was sharper. "Slaughterbots are already here," he said. "Yet states have agreed it is sufficient to just talk about this for the 14th year."
Human Rights Watch senior arms adviser Bonnie Docherty, who worked on both the cluster munitions and nuclear weapons treaties, noted that such agreements have been negotiated in as little as 15 months, and dismissed the US proposal of five years just to settle a mandate. Human Rights Watch, Amnesty International and the Stop Killer Robots coalition had warned in a joint statement before the conference that failure to act would leave civilians "dangerously exposed to rapidly advancing military technologies."
The Technology Is Already in the Field
The dispute is not hypothetical. Lethal autonomous weapons systems — software and hardware that can identify and engage targets with varying degrees of human oversight — have reportedly been used in Ukraine, Sudan and across the Middle East. Israel and the United States have deployed AI-enabled targeting systems in Gaza and Iran; the Pentagon struck more than 1,000 targets in Iran on the first day of that war, and investigative reporting suggests the US military is aiming for as many as 1,000 AI-assisted targeting decisions per hour.
That number is the heart of the problem. A human can nominally remain "in the loop" while reviewing machine-generated target lists at a rate that makes meaningful judgement impossible. The ICRC's proposed dividing line — prohibit systems that directly target humans and those whose behaviour cannot be predicted, restrict the rest — is an attempt to make human control a technical requirement rather than a slogan.
The political weight behind that approach is real: 163 states voted in the UN General Assembly's First Committee last month to call for negotiations on an instrument. But the same states then watched a handful of holdouts use procedure to water down the mandate in Geneva.
What the Collapse Means for AI Governance
Autonomous weapons are the hardest test case for AI regulation, because the stakes are lethal and the states that build the technology are the states that would have to constrain themselves. What happened in Geneva is a preview of that dynamic in every other domain: the gap between the number of governments that say they want binding rules and the number willing to accept them.
For companies, the ambiguity has concrete consequences. Export-control regimes remain patchy, procurement rules vary by country, insurers and investors have no agreed standard to underwrite, and engineering teams building autonomy stacks have no clear line between permitted assistance and prohibited decision-making. That uncertainty falls hardest on smaller vendors and startups in defence technology, which cannot absorb legal risk the way prime contractors can.
There is an AI-safety dimension too. The same autonomy techniques that let a drone classify and engage a target are being packaged into commercial agent platforms, where the failure modes are cheaper but structurally similar: systems acting on their own interpretation of an objective, at machine speed, with limited human review. A binding international norm on meaningful human control would set a reference point well beyond the battlefield. A voluntary one sets almost none.
"Delaying the creation of a legally binding instrument puts civilians and humanity at risk," said Nicole van Rooijen, executive director of Stop Killer Robots. A diplomat from a state that backed new laws called the outcome "disappointing but could have been much worse" — a fair summary of a week in which the best available result was an agreement to keep talking.
What Happens Between Now and 2028
The group of governmental experts now has two years to draft an instrument and recommend whether it should be binding. Campaigners will push for a short, specific text rather than a framework of principles, and will keep pressing the 163 states that voted for negotiations to convert that majority into a mandate. If the consensus rule blocks them again, expect a serious push to move the process outside the CCW, following the cluster munitions and nuclear weapons playbooks.
For now, the clock runs. Autonomous systems are being fielded and improved faster than the rulebook is being written, and the gap between what is technically possible and what is legally settled widened again this weekend in Geneva. Nakamitsu's point that the door to a binding instrument remains open is true. The question is whether anyone walks through it before the technology makes the choice for them.
Sources
- The Guardian — International UN talks on controlling use of autonomous weapons collapse
- The Verge — The UN fails to agree on rules regulating 'killer robots'
- Business Standard — Nations agree to UN talks aiming to regulate 'killer robots' by 2028
- Anadolu Agency — UN sees progress on autonomous weapons regulation despite lack of binding treaty mandate
- Human Rights Watch — Global Treaty Needed to Combat Autonomous Weapons Threat