There is a room where a decision exists only if everyone agrees. For twelve years that room has been deciding whether a machine may choose who to shoot. To stop the whole world there, one raised hand is enough.
Every international question has to be sent somewhere, and the address comes with a procedure attached. This one went to a group of experts at a conventional weapons convention, and that group works by consensus — not a majority, not two thirds. A decision exists only if nobody says no.
For most subjects that is sane. It stops large states outvoting small ones, and whatever comes out has everyone behind it, so it actually gets done. Now apply it here. The states refusing are few, the states asking are many: the many need the few, and the few need nobody.
Talks began in 2014. In 2023 the group was given a fresh mandate — to consider possible elements of an instrument, not to draft one. That year one delegation blocked the group's own rules of procedure, costing a full working day before the meeting could start.
Twelve years, no text, no negotiation, and no vote lost: there is nothing to point at and call a rejection, because nothing was ever put. Once, in 2024, most of the world did vote — to move the subject to the General Assembly, where a majority is enough. Nothing was put there either.
Voting no is expensive. You have to name what you object to, find allies because one vote is one vote, and your name goes into the record.
Refusing consensus costs none of that. No argument, because nothing is being tested against it. No allies, because one is enough. No explanation, because the meeting simply ends without a text — and a meeting with no text is not news. One side needs 127 agreements. The other needs one refusal.
The obvious fix is to change the procedure. But the procedure can only be changed by the procedure, and the procedure runs on consensus: the states protected by unanimity would have to agree to give up unanimity. A room built not to decide cannot decide to rebuild itself.
Nobody is cheating. Every delegate follows the rules exactly as written, and the rules were written for good reasons by people who had never heard of this weapon.
The six are the United States, Russia, Israel, India, Australia and South Korea — a list that matches no country's domestic politics, which is the first clue this is not left against right. China is in neither camp: it says yes to a binding law once the definitions are agreed, and the definitions are the thing that never gets agreed.
Their case is simple. A treaty binds whoever signs it. The states most likely to build these weapons are the least likely to sign, so a ban ties the hands of countries that were never going to build them and hands an advantage to anyone who stays outside. From inside that logic, blocking is not obstruction — it is refusing to disarm alone. Remember it: it breaks all three ways out.
Stop honouring unanimity. In November a review conference meets in Geneva and could decide by majority instead; 127 against 6 is not close.
Price: that treaty binds the states that were never the problem, which proves the objection you just heard using your own method. And consensus is the thing keeping those six in the room — break it and they walk out, and you are in way number two without having chosen it.
It has been done. Landmines, 1997. Cluster munitions, 2008. Both written away from the objectors, both grew huge, and millions of stockpiled mines were destroyed under them.
Then look at last year. Five European states left the landmine treaty — the three Baltic countries, then Finland, then Poland, the last exit taking effect in February. Ukraine suspended it instead. Poland, which destroyed more than a million of its own mines when it joined, now plans five or six million new ones for its eastern border; Finland has restarted production.
That is where this design stops. A rule built without the states that hold the weapons lasted a quarter of a century — and bent when a war reached the border of a country that had signed it.
Twelve years went on defining the machine: every definition either catches weapons that have existed for forty years, or misses the ones people are afraid of. So flip it. Ban what nobody can predict, and anything built to attack people; for the rest, fix the limits in advance — where it may operate, for how long, what it may attack — and put a human above it who can watch it work and switch it off.
Price: nobody can say what makes that human control real. In actual targeting the decision is spread across many people and many minutes, so there is no single moment to pin responsibility to, and an officer approving a hundred targets a minute meets the words exactly. And you cannot check: missile silos can be counted from orbit, but you cannot look inside the program that picks the target.
So build the checking instead of the treaty: shared test procedures, mandatory incident reports, a black box in the weapon recording what it decided and when. Now aim the same argument at that. A black box goes into a weapon because the state building it agreed — so you would get recordings from the states that were never the worry, and nothing from the six. Every road out runs through the consent of the states you most want bound.
The room is not the villain. Being outvoted is a real harm, and small states have sheltered behind consensus for decades. Everyone in that hall is following the rules correctly. That is the point.
So when something obvious does not happen, the useful question is rarely who is against it. It is which room it was sent to, and what that room needs in order to say yes. If the answer is everyone, the outcome was settled before anyone spoke.