When the Machine Decides Who Dies: Where Does the Human Stand?

On 1 December 2025 the UN General Assembly voted on a resolution concerning "lethal autonomous weapons systems": 164 states supported it, 6 opposed and 7 abstained. States seldom gather around anything in such numbers, and that alone shows that the concern is no longer confined to laboratories and science-fiction films.

What is meant by an autonomous weapon?

The idea is simpler than the headlines suggest: a system is activated after a human has set a general frame for it, and then selects and attacks its target itself, without a human ordering each strike. Researchers speak of degrees of autonomy; the human may be the one who aims and authorises, or may supervise and keep a right of veto, or may leave the moment-to-moment decision altogether. The boundaries between these degrees are not sharp, and that is where the difficulty lies.

The danger the films do not show

No conscious machine that hates humans is required. It is enough to assign a system a task in murky circumstances and for it to err as statistical systems do: misclassifying, or meeting a situation it was not trained for. In war, systems are pushed towards delegation because speed is an advantage, and whoever pauses to consult a human may think he has lost to someone who did not. Everyone may then find themselves pressed to reduce the human role even though no one wants to. This is an analysis of incentives, not a judgement about what will happen.

Then comes the problem of responsibility. If the system strikes someone who should not have been struck, who is answerable? The programmer who wrote the code years earlier? The commander who deployed it? The manufacturer? International humanitarian law was built on the premise that there is an actor who answers, and where that actor disappears, justice is lost among the parties. Our Sharia affirms the sanctity of blood and the protection of those who do not fight, and that killing is an act for which its doer is held to account; this cannot stand alongside an act for which nobody bears responsibility.

What does the ICRC recommend?

In its position of 12 May 2021, the Committee called on states to adopt new binding legal rules built on three pillars:

• Prohibit unpredictable systems: those whose effects cannot be sufficiently understood, predicted and explained, because of their indiscriminate effects.
• Prohibit systems that target people: on ethical grounds and to protect civilians and those unable to fight.
• Restrict the rest: by limiting target types, duration and geographic scope of use, and by requiring effective human supervision with the ability to intervene and deactivate in time.

Note that the first pillar ties ethics to explainability: what cannot be explained is hard to control, and hard to hold its makers to account for (see the article "Responsible Artificial Intelligence: Four Principles for Governance and Trust").

Where has the international process reached?

The file is now handled under the Convention on Certain Conventional Weapons (CCW) in Geneva, where a group of governmental experts works on a rolling text. In September 2025, 42 states stated jointly that they were ready to move to negotiating an instrument based on that text. General Assembly Resolution 80/57 then urged the parties to complete the elements of the instrument in preparation for later negotiations.

In fairness, the resolution set no binding date for negotiations, and it did not itself adopt the "two-tier" approach (prohibitions and regulation) called for by the UN Secretary-General, merely noting it in its preamble. Two sessions of the group were scheduled for 2026 (2–6 March, and 31 August–4 September), followed by the Seventh Review Conference of the Convention on 16–20 November. What it all leads to, only God knows, and it would be wrong to assert it.

What we can hold to

Beneath the technical and political disagreements lies a criterion on which people can agree: real human control over every decision that takes a life, allowing intervention and deactivation, and not a mere formal stamp. Alongside it come transparency that permits a system's decisions to be reviewed and explained, and governance agreed before deployment rather than after. Whoever possesses this technology would do well not to ask only "What can these weapons do?" but also "What should they never do, however capable they are?"

There is also an invitation to universities and research centres in our region to take part in drafting the rules rather than receiving them ready-made, and for an engineer, a lawyer, a jurist and a physician to sit together in the discussion. The technology we build is a trust, and the heaviest trust a person can be given is the blood of other people.