Foreseeing the Unforeseeable: How U.S. Negligence Law Should Address the Foreseeability of Harms Caused by Autonomous AI Agents

Abstract

As AI systems increasingly perform tasks with limited human oversight, courts will soon be required to determine how negligence law should respond when autonomous AI agents cause personal injury. Traditional foreseeability doctrine, as many scholars have observed, may fail to account for the opacity and unpredictability that characterize these systems. The main challenge could arrive when AI developers claim that a harmful outcome was unforeseeable because the specific causal pathway was novel, complex, or obscure. This article argues that such reasoning misallocates responsibility. Building on recent scholarly work, it takes the view that opacity and unpredictability are not inherent features of advanced AI systems. Rather, they are the result of abstraction choices made by AI developers, who often prioritize accuracy and efficiency over interpretability and predictability. When those choices increase the likelihood of opaque or unexpected outcomes, the legal framework should be adjusted to reflect that responsibility. In particular, where the foreseeability standard in negligence law typically makes it difficult for plaintiffs to prevail, it should be relaxed in their favor. The article examines how U.S. courts apply foreseeability across duty, breach, and proximate cause, and identifies the duty stage as the most urgent point for reform. It proposes a three-part doctrinal framework for cases involving personal injury caused by autonomous AI agents. First, courts should preserve existing law where foreseeability is already sympathetic to plaintiffs. Second, they should replace overly fact-intensive duty inquiries with clear, plaintiff-friendly categorical reasoning. Third, they should retain fact-intensive analysis at the breach and proximate cause stages to prevent overextension of liability. This approach maintains foreseeability as a meaningful constraint while calibrating it to the distinctive risks posed by autonomous AI agents.