Optimal Medical Liability for AI: Designing Legal Responsibility in AI-Driven Healthcare

Alex Chan’s 2026 working paper explores how medical liability should be structured when artificial intelligence functions as an active healthcare provider rather than simply a decision-support tool. The study focuses on the importance of the legally usable medical record, which includes patient information, AI-generated recommendations, warnings, prescriptions, follow-up instructions, and outcomes. These records determine how responsibility can be assigned by courts, insurers, regulators, and healthcare organizations.

Chan argues that AI liability is fundamentally an institutional design challenge because legal systems often operate with incomplete information. When medical records clearly distinguish errors made by AI from factors such as patient nonadherence or the natural progression of disease, traditional fault-based liability can work effectively. However, when records are incomplete or ambiguous, achieving ideal outcomes becomes difficult because liability mechanisms may unintentionally influence patient behavior.

The paper also addresses situations involving multiple contributing causes, proposing the use of marginal-responsibility scores instead of simple determinations of causation. Depending on the quality of available information, different legal approaches may be optimal, including no liability, strict liability, negligence standards, safe harbors, comparative fault, or continuous warranties. The framework further highlights how AI developers can shape future liability through the design of algorithms and record-keeping systems, emphasizing that no single legal approach is suitable for all circumstances.