Levels of Autonomy: Liability in the age of AI Agents

Lisa Soder, Julia Smakman, Connor Dunlop, Weiwei Pan, Siddharth Swaroop, Noam Kolt

Track: Sociotechnical

Keywords: AI, Autonomy, Liability

Abstract: The prospect of AI agents—autonomous systems capable of independently executing complex open-ended tasks with only limited human involvement—presents significant challenges for liability law. As AI agents become more capable, exhibit greater autonomy and act in increasingly complex environments, new questions arise regarding the attribution of liability if and when these systems cause harm. Drawing on existing governance regimes for autonomous vehicles, we propose an "autonomy scale" for AI agents to help structure legal discussions on the allocation of liability. Specifically, we analyse how key concepts in tort liability may apply to AI agents and use the UK’s Automated Vehicles Act 2024 to illustrate how an autonomy scale could be implemented in practice. This preliminary analysis, we suggest, is a useful first step in developing legal categories of AI agents and establishing appropriate liability frameworks.

Submission Number: 66