- AI in healthcare is advancing from assistive tools to autonomous decision-makers, raising patient-safety risks and liability issues.
- Current medical liability frameworks are inadequate for AI systems that use black-box reasoning, making it hard to assign blame when harm occurs.
- Clear liability frameworks are needed, modeled after graded levels used in aviation and self-driving cars, based on autonomy, automation, and operational scope.
- Seven levels of medical AI capabilities are proposed, ranging from Level 0 (informational) to Level 4 (AI-triggered human oversight), each with specific regulatory and accountability considerations.
- Existing tools (e.g., sepsis alerts) are at low levels, but future systems may operate with minimal human input, complicating responsibility assignments.