Position: AI Agents & Liability -- Mapping Insights from ML and HCI Research to Policy

The potential of increasingly autonomous AI agents and the consequent reduction in human oversight pose significant challenges to our legal system, particularly in the realm of tort law. This paper addresses the complexities of applying traditional fault-based liability frameworks to AI agents, focusing on two critical questions: "Who is best able to prevent harms?" and "What constitutes a reasonable duty of care?" We argue that by examining key properties of AI systems—namely autonomy, complexity, and adaptability—we can map existing (socio-)technical research to explicit categories of "foreseeable harms" in tort liability, as well as point to "reasonable actions" that developers can take to mitigate harms.