Three years ago, the UN Secretary-General and the head of the Red Cross gave the world a deadline. Negotiate binding rules on autonomous weapons by 2026, they said, before machines are deciding who lives and who dies without a human in the loop. That year has arrived, the rules have not, and on 25 August the two men renewed the appeal in sharper language than before.

"We are now dangerously close to crossing a moral red line," said António Guterres: "the autonomous targeting of humans by machines." The joint statement with ICRC President Mirjana Spoljaric Egger put it plainly — "innovation must serve humanity – not endanger it."

What they are actually asking for

The request is narrower and more concrete than a blanket ban. Guterres and Spoljaric Egger want governments to negotiate a legally binding instrument that does two specific things: prohibit unpredictable autonomous weapons — systems whose behaviour cannot be reliably anticipated before they are used — and prohibit anti-personnel autonomous weapons designed to target people directly. Everything else would be restricted and kept under meaningful human control rather than outlawed.

That is a deliberately achievable ask. It does not try to ban the broad category of military AI, which no major power will accept; it draws a line around the two uses that are hardest to defend.

The clock, and the room where it stops

The deadline was never arbitrary. The seventh Review Conference of the Convention on Certain Conventional Weapons — the UN's main forum for this — meets in November, and the appeal calls it "the clearest path available" to an agreement. In other words, the decisive moment is roughly three months away, and everything the two institutions have built through General Assembly debates and the CCW's Group of Governmental Experts funnels toward that room.

The problem is the room's own arithmetic, which has not changed much. As we wrote in November, the effort has broad backing — well over a hundred states favour rules — but the CCW works by consensus, and a handful of militarily significant countries have been content to keep talking rather than sign anything binding. A deadline is only as real as the process meant to meet it.

Why the language got sharper

The urgency is not abstract. The appeal comes amid unconfirmed reports that fully autonomous and AI-guided drones are already being used on the battlefield — the kind of systems that select and strike targets with no person confirming the decision. The war in Ukraine has become the working demonstration that these tools are not a future problem; the technology is deployed and iterating faster than the diplomacy meant to govern it.

That is the gap the two leaders are trying to close before it closes on its own. November will show whether "meaningful human control" becomes a line in international law, or stays what it is today: a phrase everyone endorses and no one is yet bound by.