Levels of Autonomy: Liability in the age of AI Agents

The promise of AI agents—systems capable of independently executing complex, open-ended tasks—presents significant challenges to liability law. As these systems might evolve towards greater autonomy and act in complex settings, they complicate the assignment of liability when harms occur. Drawing parallels with autonomous vehicle governance, we propose an "autonomy taxonomy" for AI agents to help structure discussions around liability assignment. The paper analyses key concepts in tort liability, explores challenges specific to AI agents, and uses the UK's Automated Vehicles Act 2024 as a case study. By proposing a preliminary autonomy taxonomy for AI agents, we aim to inform future policy discussions and research on liability frameworks that can adapt to advances in AI capabilities.