AMA Issues Governance Framework for Clinical AI Agents
The American Medical Association unveiled a governance framework aimed at guaranteeing the secure and efficient deployment of AI agents within healthcare settings. Its strategy relies on four key questions to steer how the AMA handles AI agents that carry out clinical duties, covering permissible actions and accountability. The model rests on the premise that agentic AI must be supported by evidence and deliver consistently high performance to undertake clinical roles.
Assessing Clinical Use
The paper outlines the AMA’s intended oversight of agentic AI as its applications expand in clinical environments, such as ambient documentation, diagnostic procedures, and patient monitoring. These agentic systems are capable of functioning independently, handling detailed tasks beyond the reach of standard generative AI. Because they can execute multi-step processes without direct human supervision, any erroneous or prejudiced output may lead to serious patient injury before a clinician can step in.
The framework poses four considerations for evaluating clinical AI agents: Which clinical role does the AI fulfill? What evidence confirms its capability, and what are the consequences of failure? Which actions should the AI be authorized to perform, and under what circumstances? Who bears responsibility for associated risks and outcomes? “As AI capabilities advance, evidence, authority, oversight and accountability must remain aligned in service of better patient care,” the framework states.
The association emphasizes the key involvement of physicians in both creating and rolling out the technology. The freshly issued framework will steer collaborations with other bodies to guarantee that AI agents are applied safely and responsibly in patient-care-related activities. Industry consortia are wrestling with methods to mitigate the hazards of agentic AI as its adoption accelerates. The guidance arrives at a time when the market is experiencing a rapid influx of such tools.
Rapid Adoption and Oversight Gaps
In a recent poll of 250 U.S. health-care leaders, roughly 28% indicated their institutions have already implemented agentic AI solutions, while 44% reported running pilot or proof-of-concept projects. A majority of participants anticipate a “transformative” effect of agentic AI on clinical workflows. Yet 72% said these tools are sometimes introduced without approval from IT departments, exposing a supervision shortfall.
Manatt, a legal and consulting firm, notes that 47 states introduced more than 250 AI bills affecting health care in 2025. This disparate regulatory system generates uncertainty for providers seeking to embed new technologies into routine practice.
