Who's Liable When AI and Doctors Share Medical Decisions?

Artificial intelligence is becoming embedded in hospitals and clinics, shifting from passive data assistants to systems that diagnose and make treatment decisions with little human oversight. This shift exposes gaps in existing medical liability frameworks, which assume clear-cut responsibilities among clinicians, institutions, manufacturers and regulators.

Researchers propose classifying medical AI into seven levels, similar to autonomy grading used for aircraft and self-driving cars, based on three factors: autonomy (how independently a system reasons), automation (what tasks it can perform without human input), and operational scope (the boundaries within which it operates). Levels range from fully transparent, rule-based informational tools to advisory decision-support systems and supervised automation such as closed-loop insulin delivery.

As tools climb this ladder, black-box reasoning makes it harder to verify outputs and assign fault when patients are harmed, creating 'liability gaps' where no party has clearly violated a standard. This uncertainty risks discouraging hospitals from adopting genuinely useful AI, while letting vendors sidestep post-market safety monitoring obligations.

For engineers building clinical AI, the framework offers a practical vocabulary for aligning system design, documentation and human-oversight requirements with the actual level of autonomy deployed — informing everything from explainability requirements to the audit trails needed for regulatory and legal review.