By Alexis Anagnostakis
A drone loiters above a contested district. No operator watches a screen. No commander gives a final order. The system has been told what a target looks like — a heat signature, a gait, a pattern of movement — and it has been left to decide. It selects. It engages. A family of civilians is dead.
Now ask the question that international criminal law was built to answer: who is the criminal?
For most of the last century, that question had an address. A finger on a trigger, an order down a chain of command, a signature on a deportation list. International criminal law is, at its core, a law of individual responsibility. Nuremberg established the principle that crimes against international law are committed by people, not abstractions, and that punishing those people is the only way the law has teeth. Every doctrine we rely on — direct perpetration, command responsibility, aiding and abetting — assumes a human mind that chose.
Autonomous weapons threaten to leave that question without an answer. And a crime without a criminal is not a hard case. It is a hole in the law.
What we are actually talking about
The term of art is lethal autonomous weapons systems — LAWS in the disarmament shorthand. The category that matters for criminal law is narrow but real: systems that select and engage targets without meaningful human intervention at the decisive moment. Not a remotely piloted drone, where a person still decides to fire. Not a defensive system firing at incoming missiles. The systems that break the doctrine are the ones where the choice to kill a particular human being is made by software, in the field, in real time.
These are no longer hypothetical. The debate has moved out of the seminar room and onto the battlefield. Investigators and scholars now point to the use of autonomous and AI-assisted targeting in Ukraine and elsewhere as the first generation of cases the law will have to confront. The technology is arriving faster than the rules.
Why the existing doctrines strain
Start with the most intuitive theory: hold the people responsible. The trouble is that the obvious candidates each slip the net.
The programmer or manufacturer wrote code, perhaps years earlier, with no knowledge of the specific attack, no intent that any particular civilian die, and no presence on the battlefield. War crimes require a culpable mental state. A coder who never foresaw the specific killing has, on orthodox principles, no mens rea for it. Product-liability thinking from domestic law does not map cleanly onto the criminal responsibility of an individual for a war crime.
The commander is the doctrine’s great hope — and its great difficulty. Command responsibility holds a superior liable for crimes by subordinates that the superior knew or should have known about and failed to prevent or punish. It is a doctrine designed for human troops. A machine is not a subordinate in any sense the doctrine recognises. It cannot be punished. It does not “commit” a crime that a superior fails to repress. Stretching command responsibility to cover a weapon, rather than the soldiers using it, asks the doctrine to do something it was never shaped to do.
The operator who deployed the system may be the best remaining candidate — but only if a meaningful human decision can be traced to them. Where the system is genuinely autonomous, the operator may have done nothing more than authorise a mission within parameters, leaving the lethal choice to the machine. The more autonomous the weapon, the thinner the human decision, and the harder the case against any individual.
This is what the literature now calls, with some understatement, the accountability gap. When a machine selects and attacks independently, our existing rules struggle to fix responsibility on any specific person. Everyone is in the room, and no one is holding the gun.
A weapon is not an excuse
It would be a serious error to read the accountability gap as a verdict of helplessness. It is not. The law is not silent here — it is unsettled, which is a different thing, and the direction of travel is reasonably clear.
The foundational point bears stating plainly: there is nothing about autonomy that places a weapon beyond international humanitarian law. An autonomous system is still a weapon, and weapons must comply with the cardinal principles — distinction between combatants and civilians, proportionality, the prohibition on indiscriminate attacks — in every armed conflict. A system that cannot reliably distinguish a civilian from a combatant, or that causes foreseeably excessive civilian harm, is being used unlawfully. The novelty of the means does not launder the illegality of the result.
And the dictum from Nuremberg holds: the use of novel technology to commit mass murder or to devastate civilian infrastructure does not change the calculus of command responsibility. A general who deploys a system he knows cannot discriminate, into an environment where he knows civilians are present, has made a human decision with a human mental state. The machine is the instrument. He is the author. The harder cases are real, but they should not be allowed to obscure the many cases where a human choice remains entirely visible.
The treaty that may or may not arrive
This is the backdrop to one of the more consequential disarmament negotiations of our time — and 2026 is its hinge year.
The UN Secretary-General and the President of the International Committee of the Red Cross have jointly called on states to conclude a legally binding instrument on autonomous weapons by the end of 2026. The favoured architecture is a two-tier approach: outright prohibition of the systems that can never be used lawfully — those operating without meaningful human control, or designed to target people directly — coupled with regulation of the rest to keep a human meaningfully in the loop. That approach now commands the support of well over a hundred states.
The forum is the Convention on Certain Conventional Weapons in Geneva, where a Group of Governmental Experts has worked on the problem for roughly a decade. The Group meets again in 2026 — sessions in March and around the turn of September — and is to report to the CCW’s Seventh Review Conference in November 2026. At the last General Assembly, an overwhelming majority of states backed continued movement, and a substantial bloc declared themselves ready to begin actual negotiations on the basis of the existing “rolling text.”
The obstacle is the obstacle it has always been: consensus. The CCW works by consensus, and a handful of heavily militarised states have used that procedural reality to slow the process for years. Whether 2026 produces a treaty, a framework, or another deferral is genuinely uncertain. What is not uncertain is that the technology will not wait for the diplomats.
What is really being decided
It is tempting to file all of this under arms control — a matter for disarmament specialists and Geneva delegations. That framing misses what is at stake.
The deepest commitment of international criminal law is that someone is always answerable. That a death in war is never simply a malfunction, a regrettable output, an unowned event. Strip away the human decision at the moment of killing and you do not merely create a technical lacuna in the doctrine. You attack the premise the whole edifice rests on — that accountability cannot be transferred to a machine, because a machine cannot be a defendant, cannot be deterred, and cannot be made to answer for what it has done.
The question, then, is not really whether software can comply with the rules of distinction and proportionality. It is whether we are prepared to allow a category of killing that no person is ever called to answer for. That is a choice about the kind of law we intend to keep — and it is being made, right now, in the space between what the engineers can build and what the lawyers can hold responsible.
A weapon that decides who dies, and a law that cannot decide who is to blame, cannot coexist for long. One of them will have to give. The work of this decade is to make sure it is not the law.
Alexis Anagnostakis is a criminal defence barrister practising in Athens, with more than twenty-five years at the intersection of European criminal defence, human rights, and digital evidence. He chairs the Human Rights Committee of the European Criminal Bar Association, represents Greece on the Criminal Law and Human Rights Committees of the CCBE, and is admitted to the ICC List of Counsel. He has litigated before the European Court of Human Rights, including in the Grand Chamber, and writes regularly on encrypted-communications evidence, algorithmic and AI transparency in justice systems, and the independence of the legal profession.
