Product liability for defective AI

Abstract This paper studies the efficient definition of product defects for AI systems with autonomous capabilities. It argues that defining defects in product liability law is central to distributing responsibility between producers and users. The paper proposes aligning the standard for defect with the relative control over and awareness of product risk possessed by the producer and the user. AI systems disrupt the traditional balance of control and risk awareness between users and producers. The paper provides suggestions for defining AI product defects in a way that promotes an efficient allocation of liability in AI-related accidents. It assesses whether the recent EU policy proposal on product liability aligns with this approach.