American Medical Association Outlines Liability Framework for Clinical AI

The American Medical Association has released a new liability framework addressing accountability when clinical artificial intelligence systems fail or cause harm in healthcare settings. Developed through the organization’s Center for Digital Health and AI, the framework attempts to resolve liability questions that current legal structures reportedly leave unanswered.

Addressing Gaps in Traditional Liability

According to the AMA, existing liability frameworks create significant challenges for both patients and practitioners. The organization stated that current structures could leave physicians vulnerable to liabilities for risks outside their reasonable control, while patients might face barriers when seeking recourse if an AI tool leads to adverse outcomes. Additional concerns highlighted in the framework include ambiguity surrounding product liability applications to artificial intelligence, developer contract terms and disclaimers that distance responsibility from risk control, and uncertainty regarding whether deploying healthcare organizations maintain specific responsibilities such as local testing and outcomes monitoring.

Aligning Accountability With Risk Control

The newly advanced framework establishes that multiple parties can bear liability when clinical AI errors occur. Rather than placing the full burden on a single entity, accountability is aligned based on each party’s specific ability to prevent harm. The AMA noted that while responsibility is distributed in areas of shared control, obligations do not become interchangeable, and no single party’s responsibilities disappear simply because another entity also holds accountability.

• Developers maintain obligations regarding system design, evidence, and appropriate conditions for use.
• Healthcare organizations hold responsibilities related to local testing, deployment conditions, and outcomes monitoring.
• Physicians remain responsible for clinical judgments they can meaningfully exercise based on intended use, known limitations, and available point-of-care information.

Protections and Authority for Physicians

Under the proposed framework, physician oversight is explicitly defined as something that cannot substitute for adequate evidence, proper system design, or the foundational conditions necessary to make clinical reviews meaningful. Furthermore, the AMA asserts that physicians must retain the authority to decline unproven artificial intelligence systems or tools whose clinical validation and standard of care placement remain uncertain, and they should be able to do so without facing penalties from their employing organizations.

Dr. Willie Underwood III, president of the AMA, emphasized that patients require clear avenues for recourse when clinical AI causes harm, while physicians need confidence that responsibility aligns directly with what they can reasonably know and control. The liability framework builds upon a broader clinical AI governance framework previously introduced by the AMA and its Center to guide stakeholders through issues involving regulation and technological licensure.