Guterres demands a binding ban on 'killer robots' as Geneva AI dialogue opens

One speech in Geneva on 6 July 2026, and a number that has haunted disarmament diplomats for more than a decade is back on the table: zero. Zero binding international rules govern weapons that find, decide and kill without a human in the loop. Standing at the opening of the United Nations' first Global Dialogue on AI Governance, Secretary-General António Guterres put the full moral weight of his office behind closing that gap. ²³

His framing was unusually blunt. Machines that select and engage targets and take a life without human control and judgment, he said, are morally repugnant, politically unacceptable, and must be banned by international law. ⁹ It was the kind of language the UN chief's office usually reserves for landmines or chemical weapons, and it landed in a week that has already turned AI safety into a frontline political fight inside Washington.

Guterres used the Geneva platform to do two things at once. He unveiled a new AI Child Safety Pledge alongside the autonomous-weapons appeal, recasting the debate as one that touches every household, not just ministries of defence. ⁴ And he set the stage against a clock he himself had started. Three years ago, his New Agenda for Peace named 2026 as the target year for a legally binding instrument on lethal autonomous weapons. That deadline expired, in his words, the day this dialogue opened. ⁶

The context matters. Talks under the UN Convention on Certain Conventional Weapons, the same treaty that once banned blinding lasers, have stalled for 12 years, consumed by consensus rules that any single state can weaponise to slow progress. ⁶ The Group of Governmental Experts has run out the clock on one mandate; a review conference will follow later in the year. Into that vacuum, Guterres pushed a warning that AI is advancing at runaway speed, that an experiment is being run on our own societies without a plan and without consent, and that this is not sustainable. ⁵²

What he did not say, in so many words, was who the holdouts are. He did not have to. The diplomatic picture is well known to anyone who has watched the CCW grind: a handful of military powers with operational or near-operational autonomous weapons on order or in service, and a longer list of states that would rather keep talking than start negotiating. Geneva is where those two currents meet.

The reason this debate cannot stay in committee rooms is that the technology the diplomats are being asked to govern has already been used. Reporting from the opening session noted that killer robots are no longer a hypothetical, framing the Geneva gathering as a response to a reality rather than a forecast. ³ In other words, the genie the UN is trying to draft rules for is not in a lab. It is in a fleet.

That point was underscored the same week by an unusually public fight between the United States Department of Defense and the AI company Anthropic over the terms under which frontier models can be deployed in weapons systems. ⁸¹¹ The dispute is, on its surface, a procurement argument. Beneath it sits the question Guterres put to the room: who decides when a machine is allowed to pull a trigger, and what does that decision look like on paper before it is made in code.

The official UN position is now indistinguishable from the campaign language of civil-society coalitions that have spent a decade pressing for a ban. That alignment is part of the story. Guterres did not merely endorse the goal of a treaty; he demanded that the goal be made legally binding, and he did so in the chamber where the next round of CCW talks will be measured. ²⁴

The gap is procedural. The Convention on Certain Conventional Weapons operates by consensus. A ban on autonomous weapons requires every state at the table to agree, or at least to refrain from blocking. That has not happened. It has not come close to happening. And the review conference later this year will arrive without a Group of Governmental Experts mandate still in force, a window of months rather than years during which formal treaty negotiations could be launched. ⁶