The American Medical Association released a brief outlining that doctors, AI developers and health-care organizations each hold separate duties to keep clinical AI tools from harming patients.
Distinct duties for each stakeholder
Clinicians must apply their own judgment when using AI, developers are tasked with building safe systems, and deployers are expected to pick suitable tools, train users and watch performance over time.
Jennifer Goldsack, head of the AMA Center on Digital Health and AI, warned that clinicians are being asked to act as a safety net for AI failures, exposing them to risks they may not be able to spot.
Clinical AI tools are spreading to support decision-making and cut administrative load, yet liability worries remain high among practitioners, according to an AMA survey that found a third of respondents view clear liability rules as a key adoption factor.
The new plan aims to set a clear rubric so each party knows how to reduce risk, Goldsack said.
Under the plan, clinicians must question AI outputs and can override recommendations, but they should not be blamed for errors beyond their control.
For example, if a doctor follows an AI-driven suggestion that omits critical data and the error is invisible, the doctor should not be held liable for any resulting patient harm.
Vendors must ensure their products deliver accurate advice and give users access to intended-use details, performance metrics and known limits, rather than relying solely on warnings.
Goldsack emphasized that simple disclaimers do not adequately protect patients; the plan calls for pinpointing where risk lies, assigning responsibility, and providing the necessary information to each party.
Training gaps and monitoring duties
Organizations that roll out AI, such as hospitals, health systems and physician groups, must evaluate tools before clinicians use them and supply proper education.
The 2026 AMA survey revealed that 25% of practitioners received no AI training, while 92% expressed interest in additional instruction.
Beyond education, these organizations must track tool performance to catch accuracy declines and establish procedures for handling adverse events.
Provider groups, including the AMA, have filed comments with the Food and Drug Administration, urging that AI developers should be responsible for monitoring healthcare devices that use generative AI, arguing that asking providers to shoulder that burden would place undue strain on them, particularly small and rural provider organizations.
