Abstract

A legal obligation to conduct weapons reviews is a means by which the international community can ensure that States assess whether the use of new types of weapons in armed conflict would raise humanitarian concerns. The use of artificial intelligence in weapon systems greatly complicates the process of conducting reviews, particularly where a weapon system is capable of continuing to ‘learn’ on its own after being deployed on the battlefield. This paper surveys current understandings of the weapons review challenges presented by in situ learning and outlines some concepts which may inform appropriate changes to existing weapons review processes. The main arguments made are that: a distinction should be drawn between learning capabilities that are limited to optimising existing behaviour and capabilities that extend to acquiring new behaviours; and the extent of a weapon system’s learning capabilities should be used as the basis for defining a schedule of weapons reviews.