Should a Weapon Ever Be Allowed to Choose a Human Target?
Seven Governments Want Negotiators to Draw a Legal Line Around Which Decisions a Weapon Can Make.
The first obstacle in Geneva on Monday was a dispute over who could speak. Russia objected to a rule allowing observers, including the European Union and non-governmental organisations, to take the floor. Ireland and Canada pushed back. Before delegates could get to the question of who should control an autonomous weapon, they had to settle who could address the room.
The Seventh Review Conference of the Convention on Certain Conventional Weapons opened at the Palais des Nations on 16 November 2026 and is scheduled to run until Friday. Seven governments want it to authorise negotiations on legally binding rules for weapons that can select and strike targets after a person activates them. The United States favours voluntary guidance. Russia considers negotiations on restrictions premature.
What the conference can deliver this week is a mandate. That sounds modest until you look at the competing proposals. Under one, diplomats would have two years to negotiate a binding protocol. Under another, they would spend five years working towards a non-binding agreement. Governments could leave Geneva saying they had agreed to talks under either outcome. They would have agreed to do quite different things.
What Makes a Weapon Autonomous?
Two drones could look identical from the ground. A pilot might use the first drone’s camera to identify a particular vehicle and decide whether to strike it. A person might activate the second without knowing which particular vehicle it will encounter. If the second drone can select a target and attack it on its own, it fits the International Committee of the Red Cross’s understanding of an autonomous weapon.
The distinction turns on who chooses the target. A drone can navigate by itself while a person retains the decision to use force. Conversely, a weapon can make that selection using relatively simple software. Neither the absence of a pilot on board nor the presence of artificial intelligence settles the question.
Some familiar defensive weapons already illustrate the point. A ship’s automated defences can intercept an incoming missile in the short interval before impact. The crew knows what it is trying to protect and what kind of object it expects the system to engage. That is one reason a prohibition covering every autonomous function would raise a different set of questions from a prohibition on allowing machines to select people for attack.
The Hardest Case Is a Person Who Surrenders
Consider a soldier who puts down a rifle and clearly surrenders. The person is still wearing the same uniform. The weapon may still be beside them. But the surrender changes what an opposing force may lawfully do. A system could identify the uniform and rifle correctly and still get the decision that matters disastrously wrong.
This is where the Geneva debate reaches beyond technical accuracy. Better recognition might help a machine distinguish a person from an object. Whether that person can lawfully be attacked depends on their conduct and circumstances at the time. Those circumstances may have changed since anyone last saw them.
International humanitarian law already governs attacks carried out with autonomous weapons. Parties to a conflict must distinguish civilians from lawful military targets, take feasible precautions to protect civilians and refrain from attacks expected to cause civilian harm excessive in relation to the concrete and direct military advantage anticipated. A commander cannot hand those obligations to a machine.
The difficult question comes before activation. What can the person authorising force reasonably know about the circumstances in which the weapon will act? On a ship facing an incoming missile, the relevant circumstances may be apparent and short-lived. Where people are moving, surrendering or coming to someone’s aid, a judgement made earlier can lose its basis while the weapon continues to operate.
What Would Binding Rules Actually Restrict?
The Red Cross proposes two prohibitions. It wants governments to rule out autonomous weapons whose effects people cannot adequately understand, predict and explain. It also wants a prohibition on systems designed or used to apply force autonomously against people. For other autonomous weapons, it advocates restrictions on where, for how long and under what human supervision they can operate.
Those proposals answer different concerns. The first addresses what happens when even the people authorising a weapon cannot sufficiently anticipate its effects. The second draws an ethical line that would survive improvements in accuracy. Under the Red Cross proposal, a manufacturer could demonstrate that a system was exceptionally good at recognising individuals and still leave the fundamental objection unanswered. Should recognition by a machine ever be enough to decide that a particular person may be killed?
Rules of that kind would give armed forces an answer while they were still deciding which capabilities to acquire. Existing law can require a commander to judge whether a particular use would be lawful. A specific prohibition could settle an earlier question for a state that accepted it, by ruling out an entire use in advance. The Red Cross has set out what it wants those rules to say. Governments have yet to agree even to negotiate them.
Why Britain, America and the Treaty’s Backers Want Different Mandates
Austria, Belgium, Brazil, Ireland, Mexico, New Zealand and Panama submitted a proposal in October to begin negotiating a legally binding agreement. The conference chair has put forward a two-year mandate to negotiate a protocol. Both approaches would allow states to argue over the detail while giving negotiators a clear legal destination.
Washington’s five-year proposal points towards non-binding guidance. Its argument starts from an important fact shared by the other side. The laws of war already apply. Guidance could help armed forces interpret those duties as technology changes, and governments could put it into practice without waiting for a new treaty to enter into force. Its weakness is that a state could endorse a principle and still dispute what, precisely, it had committed itself to do.
Britain, France and Germany favour starting negotiations while leaving the legal form open, according to reporting from Geneva. That could bring governments to the table that would reject an explicit treaty mandate. It would also defer the very disagreement the seven sponsors want settled. Years of work could produce a text whose substance most delegations accepted, only for them to divide again over whether anyone should be legally bound by it.
The convention works by consensus. Even broad support for a treaty cannot carry the decision over a sustained objection. The price of keeping major military powers involved may therefore appear in a single phrase describing what the talks are for.
What to Look for When the Conference Ends
Three details will matter in any closing text on Friday. Does it instruct diplomats to negotiate an agreement? Does it specify that the result should be legally binding? And does it give them a deadline? A statement can express concern about autonomous weapons while leaving all three questions open.
Even the strongest mandate on offer would leave substantial work ahead. Negotiators would still need to define which systems and uses fell within the rules, agree where prohibitions should apply and persuade states to accept the resulting obligations. Nothing agreed about a negotiating timetable would, by itself, prohibit a weapon on Saturday morning.
Yet the wording matters to someone far removed from the conference hall. Return to the soldier who has put down a rifle. Under one approach, each state would decide how its existing legal duties apply to a machine encountering that person. Under the Red Cross proposal, states accepting the prohibition would have made a prior commitment that a machine could never select that person for attack. The choice facing Geneva this week is whether to start writing a rule capable of giving that soldier the same answer across borders.
Sources
United Nations — Seventh Review Conference programme and dates
Reuters — US and Russia to oppose talks on limiting autonomous weapons
Associated Press — Geneva conference considers negotiations on autonomous weapons
International Committee of the Red Cross — Position on autonomous weapon systems
International Committee of the Red Cross — Autonomous weapons and the laws governing their use
Next Reads
Russia Issues Chilling Warning To Britain Over Drones Striking Deep Inside Its Territory — Why supplying a weapon and controlling its use raise different questions for Britain and Ukraine.
Elon Musk’s SpaceX Cuts Starlink Access as Ukrainian Drone Missions Fail and Kyiv Fears Trump Pressure — What happens when a military depends on a communications network controlled by a private company.
What If Ukraine Wins the War? — How drones, industrial capacity and security guarantees shape competing definitions of victory.