System Architects and Integrators as ‘Most Responsible’ for Unlawful Attacks? Re-centring Article 25(3)(a) in Autonomous Weapons Campaigns

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Debates about autonomous weapon systems (‘AWS’) usually begin with the commander, the operator, and the immediate moment of attack. That frame is necessary, but it is not sufficient. In AWS-centred campaigns, decisive choices for civilian protection may be made earlier, when system architects, model leads, integrators, and authorised risk owners determine how a system will classify persons, objects, places, and patterns of life. This article argues that a narrow class of design-stage actors may, on the proper facts, be treated as potential principals under Article 25(3)(a) of the Rome Statute of the International Criminal Court (‘Rome Statute’). The claim is limited. Engineers are not liable as a class, and international criminal law cannot provide a general regime for military artificial intelligence. The relevant case is one in which a small group has practical authority over design choke-points, knows that the configuration will make unlawful harm occur in the ordinary course of a campaign, and still maintains that configuration. That case matters because Article 25(3) of the Rome Statute begins with the proposition that ‘[i]n accordance with this Statute, a person shall be criminally responsible and liable for punishment for a crime within the jurisdiction of the Court if that person’ 1 and then assigns liability through the modes that follow. This article makes three linked points. First, system architects and integrators should be understood by function, not by job title. System architects shape the architecture, thresholds, safeguards, logs, and release gates of a weapon system; integrators adapt that system to a theatre, platform, command structure, and operational concept. They may work in private firms, military laboratories, procurement bodies, or mixed programme offices. Secondly, AWS should be kept separate from AI-enabled decision support systems (‘AI DSS’). A system that ranks or recommends targets raises one legal issue. A system that selects and applies force after activation raises another. Thirdly, the many-hands problem should be treated as a problem of proof and individual attribution in criminal law. It should not be used as a reason to avoid individual criminal responsibility. Design authority, procurement review, foreseeability, opacity, and mens rea are therefore examined through the existing control theory of co-perpetration.