What the law of armed conflict actually settles about autonomous targeting — and what it genuinely leaves open
International humanitarian law was not written for machines that choose their own targets. Its foundational instruments — the Geneva Conventions of 1949 and their Additional Protocols of 1977 — assume a human being at the point of decision: a commander who distinguishes a combatant from a civilian, who weighs an anticipated military advantage against expected civilian harm, and who can be held responsible for getting it wrong. Autonomous weapon systems trouble every one of those assumptions, not by breaking the rules but by removing the human from the moment the rules were designed to govern. The result is a legal debate that generates a great deal of heat and, on inspection, a surprisingly clear division between what the law already settles and what it genuinely leaves open.
This Signal is an attempt to draw that line honestly. Much of the public argument about autonomous weapons treats every question as equally unresolved, which serves advocacy on both sides and clarity on neither. The reality is more useful. Three things the law already says with confidence. Three things it has not decided, and where the current treaty process is trying to decide them. Keeping those two lists distinct is the whole of the analytical task, and it is worth doing because the people who most need the distinction — procurement officials, weapons reviewers, operational lawyers — are being handed a debate that blurs it.
Three things the law already settles
Begin with what is not in doubt, because it is more than the debate usually admits. The first settled point is that international humanitarian law applies to autonomous weapon systems in full. There is no autonomy exception. In 2019 the High Contracting Parties to the Convention on Certain Conventional Weapons formally reaffirmed that IHL applies to the development and use of such systems, and the UN General Assembly echoed that position in resolutions adopted in 2023 and again in 2024. The rules of distinction, proportionality, and precautions in attack bind an autonomous system exactly as they bind any other means or method of warfare. A weapon that cannot comply with those rules is unlawful, whatever its degree of autonomy. This is not contested by any serious state party.
Epistemic status: Confirmed. The applicability of existing IHL to autonomous weapon systems is affirmed in the CCW Group of Governmental Experts’ Guiding Principles (2019) and reaffirmed in UN General Assembly resolutions in 2023 and 2024. It is the common ground on which every party to the debate stands.
The second settled point is that the existing rules already prohibit certain autonomous systems outright. A system whose effects cannot be anticipated or limited — one that is, in the technical sense, unpredictable in the environment of its use — cannot satisfy the requirement of distinction, because a user who cannot foresee what the system will strike cannot ensure it strikes only lawful targets. Such a system is indiscriminate by nature, and weapons that are indiscriminate by nature are already prohibited under customary law and under Additional Protocol I. The International Committee of the Red Cross, in its October 2025 position paper, makes this point directly: existing rules, being technology-neutral, already rule out certain kinds of autonomous weapon without any need for new law. The prohibition is not aspirational; it follows from rules that are already in force.
Epistemic status: Confirmed. The prohibition on weapons that are indiscriminate by nature is established customary IHL, reflected in Additional Protocol I, Article 51(4). Its application to unpredictable autonomous systems is set out in the ICRC’s 2025 position paper and is not, as a matter of existing law, seriously disputed.
The third settled point concerns accountability. Legal responsibility for an attack cannot be delegated to a machine, because a machine cannot be a subject of international law. The obligation to comply with IHL rests on human commanders and operators, and the responsibility for an unlawful attack carried out by an autonomous system remains with the humans who deployed it. A weapon system is a means of warfare, not a combatant; it does not hold obligations and cannot be blamed. This is why the recurring phrase in the treaty debate is human responsibility rather than machine accountability — the former is a legal fact, the latter a category error.
Epistemic status: Confirmed. That legal responsibility for the use of force remains with humans is a settled feature of IHL; the SIPRI–ICRC work on retaining human responsibility in the development and use of autonomous weapon systems sets out the principle in detail.
Three things the law genuinely leaves open
Now the harder list, and the honest one. If the settled points are more than the debate admits, the open points are also more real than the reassurance from weapons developers suggests. These are not questions that existing doctrine quietly answers if one reads carefully. They are genuine gaps, and the current treaty process exists precisely because states have not agreed on how to fill them.
The first open question is what degree of human control the law actually requires. Everyone agrees that some human involvement is necessary; no one has agreed on how much, of what kind, or at which point in the targeting cycle. Must a human approve each individual engagement, or is it sufficient to set the parameters within which a system then operates autonomously over some period, area, and class of target? The CCW experts’ rolling text speaks of maintaining context-appropriate human oversight, particularly at the identification and engagement of targets — but ‘context-appropriate’ is precisely the term over which states disagree. Existing IHL requires that attacks be directed by humans capable of making the necessary judgements; it does not specify the mechanism by which that direction must be exercised over a system that operates without further intervention once launched. That specification is the single most contested question in the field, and it is genuinely open.
Epistemic status: Open — a matter of active negotiation, not settled doctrine. That meaningful or context-appropriate human control is required is widely asserted; its precise legal content is undefined and is the central subject of the CCW Group of Governmental Experts’ rolling text and the parallel UN process.
The second open question is whether autonomous systems may lawfully target human beings at all, as distinct from objects. A meaningful strand of the treaty debate holds that systems which select and attack people — as opposed to materiel such as vehicles, radars, or incoming munitions — cross a line that materiel-targeting systems do not, because reducing a human being to a sensor-matched target profile is held to offend the principle of humanity and human dignity in a way that existing rules do not squarely address. Whether that distinction has a basis in current law, or would require new law to establish, is unresolved. Existing IHL distinguishes combatants from civilians; it does not, in terms, distinguish between autonomously attacking a tank and autonomously attacking the person inside it. Many states and the ICRC argue the latter should be specifically restricted; others resist. The law as it stands does not decide the matter.
Epistemic status: Open, and contested. The proposal to prohibit or restrict autonomous systems that target persons is central to the ICRC’s recommendations and to the position of a large group of states, but it is a proposal for new rules, not a settled reading of existing ones. Presenting it as either already-law or clearly-not-law would misstate the position.
The third open question is how the proportionality judgement — the weighing of anticipated military advantage against expected civilian harm — can be performed by, or delegated to, an autonomous system. Proportionality is not a calculation; it is an evaluative judgement that IHL entrusts to a reasonable commander in the specific circumstances of an attack. Whether that judgement can be encoded, whether a system can be designed to make it, or whether the requirement effectively means a human must make each proportionality assessment in something close to real time, is not resolved by existing doctrine. The rule is clear; its application to a system that acts without contemporaneous human judgement is not. This is less publicly prominent than the human-control and anti-personnel questions, but it may be the most technically intractable of the three.
Epistemic status: Open. That the proportionality rule applies is Confirmed; whether and how it can be satisfied by or through an autonomous system is genuinely unsettled, and turns on questions of context-specific human assessment that the ICRC’s 2025 paper identifies as central and unresolved.
Where this is going
The reason the open questions matter now, rather than as a matter of academic interest, is that the process meant to resolve them has a clock on it. In October 2023 the UN Secretary-General and the President of the ICRC jointly called on states to conclude a new legally binding instrument on autonomous weapons by 2026. The General Assembly has since moved the issue onto its own agenda, adopting a first resolution in December 2023 and a second in December 2024 by a vote of 166 to 3 — the three opposed being Belarus, North Korea, and Russia — with more than 120 states supporting negotiation of a new treaty. The pattern is unmistakable to anyone who has watched the landmine and cluster-munitions conventions: sustained majority pressure, resistance from a small number of militarily significant states, and a gradual migration of the issue out of the consensus-bound CCW forum, where a single state can block progress, toward forums where it cannot.
For a defence-technology reader, the practical implication is not that a ban is imminent — the major developers, the United States chief among them, favour voluntary codes over a categorical prohibition, and the outcome is genuinely uncertain. The implication is that the three open questions are the ones on which design decisions taken now will be judged later. A system architected today around a defensible answer to the human-control question, the anti-personnel question, and the proportionality question is a system whose lawfulness will survive whatever the treaty process produces. A system architected to the current ambiguity, on the assumption that what is not yet prohibited is therefore permitted, is a procurement risk. The law will move. The settled points will not, and the open points are being closed. Reading the difference correctly is the difference between building to the law’s trajectory and building to its present silence.
References
Primary legal instruments
Geneva Conventions of 12 August 1949 (I–IV), common provisions on grave breaches and enforcement.
Protocol Additional to the Geneva Conventions of 12 August 1949, and relating to the Protection of Victims of International Armed Conflicts (Protocol I), 8 June 1977 — in particular Article 48 (basic rule of distinction), Article 51 (protection of the civilian population, including Article 51(4) on indiscriminate attacks), Article 52 (protection of civilian objects), and Article 57 (precautions in attack).
ICRC Customary IHL Study (2005), Rules 1, 7, 11–14 (distinction, indiscriminate attacks, proportionality, and precautions).
Intergovernmental process and official documents
Group of Governmental Experts on Emerging Technologies in the Area of Lethal Autonomous Weapons Systems (GGE), Guiding Principles, annexed to the Final Report, CCW/MSP/2019/9, December 2019 — reaffirming that IHL applies to lethal autonomous weapons systems.
CCW GGE on LAWS, rolling text of 8 November 2024, outlining potential regulatory measures including predictability, traceability, and context-appropriate human oversight.
UN General Assembly Resolution 78/241, ‘Lethal autonomous weapons systems,’ adopted 22 December 2023 (152 in favour, 4 against, 11 abstentions).
UN General Assembly Resolution on lethal autonomous weapons systems, adopted 2 December 2024 (166 in favour, 3 against — Belarus, North Korea, Russia — 15 abstentions).
UN Secretary-General, Report on lethal autonomous weapons systems, August 2024 (reflecting submissions from more than 70 states and from the ICRC and civil society).
Institutional positions and analysis
ICRC, Autonomous Weapon Systems and International Humanitarian Law: Selected Issues — Position Paper, October 2025.
ICRC, Position on Autonomous Weapon Systems (recommending new legally binding rules, including a prohibition on unpredictable autonomous weapon systems and restrictions on systems that target persons), 2021 and subsequent updates.
Statement of ICRC President Mirjana Spoljaric to the UN informal consultations on lethal autonomous weapons systems, 12 May 2025.
Joint appeal of the UN Secretary-General and the President of the ICRC for a legally binding instrument by 2026, 5 October 2023.
- Boulanin, N. Davison, N. Goussac and M. Peldán Carlsson, Limits on Autonomy in Weapon Systems: Identifying Practical Elements of Human Control, ICRC & SIPRI, June 2020; and SIPRI, Retaining Human Responsibility in the Development and Use of Autonomous Weapon Systems, October 2022.
American Society of International Law (ASIL), ‘Lethal Autonomous Weapons Systems & International Law: Growing Momentum Towards a New International Treaty,’ Insights, 2025.
