China accelerates military humanoid robot development
There are no wounded yet from this particular development, no counted dead, no hospital ledgers to consult - and that absence is precisely what should concern us. The People’s Liberation Army and its researchers in China are building the means to inflict future casualties whose number we cannot yet name, in conflicts whose location we cannot yet specify. Article 36 of the First Additional Protocol to the Geneva Conventions requires that any new weapon, means, or method of warfare be assessed, before its use, against the obligations of international humanitarian law. The question before us is not whether humanoid combat robots will someday appear on a battlefield. It is whether that review is happening now, in Beijing, in any form subject to outside verification - and I have seen no evidence that it is.
It is a choice. When the PLA’s researchers describe the acceleration of humanoid systems for warfare, they describe capability. They do not describe the safeguards required to make that capability lawful: a targeting architecture that can distinguish combatant from civilian, wounded from able-bodied, surrendering soldier from active threat. At Solferino I watched men who had stopped fighting go on being treated as enemies simply because no one nearby recognised that the fighting, for them, was over. A soldier’s raised hands, a soldier’s collapse, a soldier’s white flag - these are signals a machine must be built to recognise before it is built to shoot. Nothing in the reporting on this programme tells us whether recognition of that kind is even part of the design specification, and the burden of proof lies with Beijing, not with the rest of us.
Here the strongest objection deserves a fair hearing. One could argue that a machine, unlike a frightened conscript, cannot panic, cannot seek revenge, cannot fire out of fear for its own life - that a well-governed autonomous system might in principle discriminate more reliably than the exhausted men I found lying in the Lombard sun in 1859. I do not dismiss this. Discipline built into steel and code, rather than demanded of a terrified nineteen-year-old, is not automatically inhumane. But the argument only holds if the discrimination is actually engineered, actually tested against real battlefield conditions, and actually reviewed by some body capable of saying no. A rule that exists only as a technical possibility is not a rule. It is a marketing claim. The Convention does not ask whether a weapon could theoretically comply with the principle of distinction; it asks whether the state deploying it has verified that it does, and whether that verification can be checked by anyone outside the programme itself.
This is where the institutional gap sits, and it is wide. The International Committee of the Red Cross has spent years pressing for international agreement on autonomous weapons systems - pressing, and largely being ignored, by every major military power, not China alone. There is no treaty body with the authority to inspect a Chinese research facility the way our delegates once moved between field hospitals checking that the wounded of both armies received equal care. There is no equivalent of the armband, the recognised emblem that tells a soldier: this is a protected person, hold your fire. A humanoid robot has no such emblem to read, and no obligation yet compels its makers to teach it one.
What is needed is not a treaty banning robots, which no state contemplating strategic advantage will sign, but a demand - from the ICRC, from other governments, from any forum where China still seeks legitimacy - that Article 36 reviews for these systems be documented and shared, at minimum with a neutral body empowered to ask hard questions. Absent that, we are not debating a new chapter in the laws of war. We are watching a new weapon walk toward the battlefield while the law waits outside, still checking its papers at a door that has not yet been built.