The AMA wants to clarify who's on the hook when clinical AI goes wrong

As AI increasingly steps into clinical decision-making, a big question looms: when something goes wrong, who's responsible? The American Medical Association (AMA) is stepping in with a new liability framework that takes a clear stance — accountability should follow the ability to prevent harm, and it should be shared across all parties involved.

The framework, released by AMA's Center for Digital Health and AI, identifies three key players: AI developers, healthcare organizations deploying the tools, and the physicians using them. Each carries distinct obligations. Developers must ensure their systems are safe and validated. Organizations must conduct local testing and monitor outcomes. And physicians are responsible for the clinical judgment they can meaningfully exercise — but only within the bounds of what they can reasonably know and control.

Importantly, the AMA draws a firm line: physician oversight cannot substitute for proper system design or adequate evidence. Doctors also retain the right to decline AI tools whose clinical validation remains uncertain — without fear of organizational penalty.

Key Takeaways:

• Liability is distributed based on each party's capacity to prevent harm — not interchangeable across parties
• Physicians cannot be penalized for opting out of unproven AI systems
• Existing liability frameworks leave critical gaps that could expose physicians to unfair risk

Why it matters: With agentic AI rapidly entering clinical workflows and federal regulation lagging behind, the AMA's framework offers a much-needed roadmap for accountability — protecting both patients seeking recourse and physicians navigating an evolving technological landscape.