Opinion: The "killer robots" talks keep failing for a reason that has nothing to do with killer robots
A diplomatic process that lets a single state veto the will of more than a hundred others was never built to regulate anything, least of all autonomous weapons.
A diplomatic conference that just concluded in Geneva illustrates a problem that has little to do with technology and everything to do with how international rules get made. For a week, representatives of 128 countries met under the Convention on Certain Conventional Weapons to decide whether to finally begin negotiating binding limits on autonomous weapons systems, machines capable of selecting and firing on targets without a human pulling the trigger. The conference, which the UN's own Geneva office describes as running Nov. 16 to 20, was the fourth such review since the issue was first raised at the UN more than a decade ago. The substantive case for regulation is no longer seriously contested. What keeps failing is the process built to act on it.
That process runs on consensus: every one of the convention's 128 member states effectively holds a veto over anything the group produces. That rule is defensible when the goal is building durable agreement among wary militaries. It is close to unworkable when the declared will of an overwhelming majority can be blocked by a handful of governments that would rather it not exist. This month's conference is only the latest demonstration of that gap, and it is worth being precise about how it has opened.
What got removed in September, and by whom
The conference's working material came from a Group of Governmental Experts that spent three years drafting "elements of an instrument" on autonomous weapons, under a mandate that asked it to proceed by consensus. Its final negotiating session ran from Aug. 31 to Sept. 5 in Geneva, and an official UN transcript of its opening meeting shows the stakes were already clear to delegates: Izumi Nakamitsu, the UN's high representative for disarmament affairs, told the room that "the window for effective international action is narrowing," while the Holy See's delegation argued flatly that "no algorithm can make war morally acceptable." Algeria, speaking for the African Group, and Brazil, speaking for roughly three dozen states, both said the draft text already provided a sufficient basis to begin binding negotiations.
According to reporting on that session's final days, American and Russian delegates used a closed late-week drafting session to remove language requiring human review of AI-identified targets before a strike, provisions on the predictability and reliability of autonomous systems, and broader references to human rights law, narrowing the text's legal grounding to humanitarian law alone. More than 70 of the convention's 128 states had backed retaining that language as a foundation for a binding protocol. Because the group operates by consensus, two delegations were able to rewrite what a clear majority had already endorsed, and nothing in the process required them to explain why.
The fair case for caution
It would be wrong to treat every objection to a fast binding treaty as bad faith. Autonomous weapons are genuinely hard to define in law: militaries already use automated defensive systems, like missile-interception batteries, that make split-second engagement decisions with limited human input, and a sloppily worded prohibition could sweep in systems nobody is actually trying to ban while leaving the dangerous ones untouched. The United States has argued, in a working paper reported on ahead of the Geneva conference, that existing international humanitarian law already applies to these weapons and that a nonbinding instrument can be revised faster than a treaty as the technology changes; it proposed five more years of expert-group study rather than a negotiating mandate. That is a coherent position, even if one disagrees with it, and a rushed treaty that major military powers simply ignore would arguably do less for civilian protection than a slower process that eventually binds them.
The trouble is that "five more years of study" is not a neutral continuation of deliberation. It is also, functionally, the policy outcome preferred by the states least eager to be bound by anything, achieved through a procedural rule rather than through persuading the other 120-odd governments that nonbinding guidance is sufficient. UN human rights chief Volker Türk made the distinction plainly in remarks ahead of the conference, arguing that "a machine must never have the power to decide whether a human being survives or dies," and Secretary-General António Guterres opened the review conference itself by urging delegates to "take action to prevent the use and spread of these merciless weapons." Those are not fringe positions within the UN system; they are the institution's own leadership describing a majority view that its own decision-making rules are structured to frustrate.
A majority that keeps finding the same ceiling
The scale of that majority is not in serious dispute. The General Assembly, which operates by vote rather than consensus, adopted Resolution 78/241 on lethal autonomous weapons systems in December 2023 by a recorded vote of 152 in favor, four against and 11 abstentions. That is as close to global consensus as the General Assembly produces on a contested security question. It is also a vote with no binding force, precisely because the instrument capable of producing binding rules, the CCW, requires unanimity to move at all. The campaign group Stop Killer Robots has put the gap in similarly stark terms, estimating that roughly 130 countries now support a legally binding instrument and more than 70 are ready to begin negotiating one, against a blocking minority of roughly a dozen that includes the United States, Russia and Israel. A rule built to prevent any state from being bound against its will has, in practice, become a rule that lets a small fraction of the world's governments decide the ceiling for what the other nine-tenths may achieve.
What actually needs fixing
This is the part of the story that gets lost when each review conference is covered as a fresh disappointment rather than a recurring structural failure. The 2021 review conference also ended without a negotiating mandate. The 2026 Group of Governmental Experts process also ended with its strongest language removed in a closed session. The pattern is not that autonomous weapons are an unusually hard problem to legislate, though they are not simple; it is that a body requiring unanimous consent among 128 governments, several of which are actively developing the weapons in question, will reliably produce the minimum any one of them will accept, regardless of where the other hundred-plus stand. States that want binding rules on autonomous weapons have a narrower set of real options than the CCW's structure suggests: pursue a separate treaty process outside the consensus framework, as was done for landmines and cluster munitions over the objections of major powers, or accept that this convention will keep producing study mandates rather than rules. Either path is more honest than treating the next review conference, five years from now, as though it might finally break a pattern this one did not.

Opinion: A Judge Was Right to Stop Washington From Deciding What Foreign Students May Study

Opinion: The Supreme Court Just Put a National Park Beyond the Law's Reach
