AMA releases governance framework for clinical AI agents

The framework calls for specifying an AI system’s clinical task, patient population, care setting and intended use.[3] The AMA lists possible functions including documentation, monitoring, triage, recommendations, diagnosis, treatment and prescribing.[3] These functions span different levels of clinical involvement: the framework describes AI authority as ranging from informing or recommending to deciding or acting.[3] For teams assessing a proposed system, the framework’s structure makes the intended role a starting point for determining what evidence and oversight are appropriate.[3]

The AMA’s governing principle is that AI clinical authority should be earned through evidence and maintained through demonstrated performance.[3] The framework says oversight and authority can change as evidence, performance, risk or technology changes.[3] Evaluation should address performance, safety, clinical benefit, population fit and uncertainty, with the degree of evaluation proportionate to potential consequences.[3] The AMA also says greater consequences and greater autonomy call for stronger evidence.[3] It describes an ongoing cycle of deploying AI, monitoring it, reassessing it and learning from its performance.[1]

The framework says accountability should reflect each actor’s authority, control, knowledge of risk and ability to prevent or mitigate harm.[3] It identifies developers, deployers, health systems, physicians, payers and other actors as participants in the governance landscape.[3] The AMA says the framework addresses liability, licensure and regulation as areas for work with providers and other stakeholders.[2] In this context, “licensure” means authorizing an AI system to operate in a defined clinical role, not licensing it to practice medicine.[1]

In the same survey, 72% said such tools are sometimes implemented without IT approval.[4] Healthcare Dive also reports that 47 states introduced more than 250 AI bills affecting healthcare in 2025, amid what it describes as a lack of substantive federal AI regulation.[4] The AMA says it will bring physicians together with AI builders, deployers and regulators to shape governance for better patient care.[3]