The American Medical Association introduced guidance asking healthcare organizations to assess agent tasks, evidence, permitted actions and accountability.
The American Medical Association has released a governance framework for AI agents that perform clinical functions, offering a set of questions intended to help guide decisions about deployment. The framework centers on four issues: what clinical function an agent performs; what evidence supports its ability to do so and what happens if it fails; what actions it should be allowed to take and under what conditions; and who is accountable for risks and outcomes.
The guidance addresses a shift from AI tools that generate information to agentic systems that can complete complex, multi-step tasks. Healthcare Dive noted that possible uses include ambient documentation, diagnostics and patient monitoring. Those applications differ in clinical risk, so the framework encourages organizations to assess the specific task and operating conditions rather than treating all AI agents as interchangeable.
The AMA’s approach emphasizes that evidence, authority, oversight and accountability should remain aligned with patient care. That is consequential in clinical settings, where an inaccurate or biased action could affect a patient before a clinician has an opportunity to intervene. The framework is guidance from a professional association, not a new federal regulation or proof that any particular agent is safe or effective.
The timing reflects uneven adoption and oversight. Healthcare Dive cited a survey of 250 U.S. healthcare leaders in which 28% said their organizations had deployed agentic AI and 44% reported pilots or proof-of-concept projects. The same survey found that 72% said such tools are sometimes implemented without IT approval. Respondents also flagged compliance concerns and unauthorized access to protected health information.
For health systems, the questions offer a practical starting point for procurement and governance reviews: define the agent’s role, establish evidence and failure procedures, set limits on autonomy, and identify who is responsible. The framework also gives the AMA a basis for collaboration with other organizations as clinical AI use expands. It does not itself resolve liability or replace applicable privacy, medical-device or professional requirements, and healthcare organizations will still need controls tailored to their own workflows and jurisdictions.