The Interval, Not the Autonomy: Why Ten Years of Defining Autonomous Weapons Produced Nothing, and What to Regulate Instead

The Group of Governmental Experts on lethal autonomous weapons systems reports to the Seventh Review Conference of the Convention on Certain Conventional Weapons in November 2026, after a decade of work. It has produced a rolling text and no instrument. This paper argues that the reason is not the disagreement usually blamed for it, and that the object the debate has organised itself around is the wrong one. The structural finding comes first, because it changes what a proposal has to do. The Group's mandate is to formulate, by consensus, a set of elements of an instrument without prejudging its nature. States therefore have no mandate to negotiate. A consensus rule combined with an express instruction not to prejudge the outcome is not a slow route to a treaty; it is a procedure with no terminus, and the decade of definitional argument is its symptom rather than its cause. Any proposal offered into the 2026 window must be capable of adoption by a body that cannot negotiate. The substantive argument is that the decisive variable is temporal, and that both halves of it are already in the debate without being joined. The rolling text's central formula, "context-appropriate human judgement and control", cannot be verified: no state can be held to it and no inspector can check it, because nothing in it specifies what would count. Within the same text sits a scale-of-operation measure, limiting the types of targets, duration, geographical scope and the number of engagements, in which duration appears as one component among four, unexplained. Outside the text, the principal gap identified by its most careful critics is the absence of any requirement that a user adequately understand a system and its effects in the context of use. This paper's claim is that those two are halves of one variable, and that the variable is not one component among four but the one determining whether the others can be satisfied at all: the interval between the last human decision and the effect, together with what was knowable across that interval. Distinction, proportionality and precautions are judgements made at a moment, on facts obtaining at that moment. The legal question is whether those facts still hold when the effect arrives. The reframing says that the lawful interval is a function of the rate at which the relevant facts decay, that this rate is a property of the situation rather than of the machine, and that it is estimable in advance and checkable afterwards. A system may lawfully loiter for a week over empty ocean and unlawfully for ninety seconds above a moving crowd. No single figure belongs in a treaty. A rule relating permitted interval to fact-decay rate does, and unlike "context-appropriate" it can be complied with, audited and breached. A second finding concerns Article 36 of Additional Protocol I: its temporal reach is broad across study, development, acquisition and adoption, yet all four stages are pre-employment. The provision has wide reach across procurement and no reach across operational life. The answer proposed is a second occasion for determination, at employment, rather than a wider scope. The framework is applied to loitering munitions, automated air and missile defence, and emplaced offline ground systems, and distinguishes cases the autonomy framing cannot. The paper closes with a drafted element in the form it would have to take to be tabled, an account of what it would cost each of the principal states to accept, and four routes to adoption, three of which require nobody's agreement.