The American Medical Association (AMA) has published a governance framework for “agentic AI” systems that take on clinical functions, built on the principle that an AI system’s clinical authority “should be earned through evidence and maintained through demonstrated performance.” The framework, posted on the AMA’s website on Oct. 13, poses four questions that the association says will guide its work on liability, licensure and regulation as AI moves from drafting notes toward making consequential clinical decisions.
Agentic AI is the industry term for software that can plan and carry out multi-step tasks toward a goal on its own, rather than producing a single output for a person to act on. In health care such tools are already used for ambient documentation, diagnostics and patient monitoring, and because they can run multi-step tasks without human oversight, an inaccurate or biased decision could cause significant patient harm before a clinician intervenes, Healthcare Dive reported.
Four questions
The AMA framework organizes its approach around four areas, each framed as a question:
- Function: “What clinical function is the AI performing?”
- Evidence and risk: “What demonstrates that it can perform that function, and what happens if it fails?”
- Authority and oversight: “What should the AI be permitted to do, and under what conditions?”
- Accountability: “Who is responsible for the risks and outcomes?”
The document is explicit that the answers are not fixed. “Authority and oversight can evolve as evidence, performance, risk or the technology itself changes,” the AMA wrote, adding that “as AI capabilities advance, evidence, authority, oversight and accountability must remain aligned in service of better patient care.” The framework was developed by the AMA’s Center for Digital Health and AI, which the association says “builds partnerships with regulators, policymakers and technology leaders to shape benchmarks for safe and effective use of AI and other digital tools in medicine.”
The framework is a short position statement and infographic rather than a detailed technical standard. It does not define tiers of autonomy, specify what evidence would qualify a system for a given function, or name any product. The AMA describes it as setting out “how the AMA is approaching the governance of agentic AI assuming clinical functions and how we will focus our work in this area.”
AMA chief executive John Whyte, M.D., said in a statement reported by Digital Health News that the growing capabilities of AI systems are raising urgent questions about clinical authority, oversight and accountability, and that the sector cannot resolve those questions through trial and error in patient care. The framework, he said, will guide how clinical authority is assigned on the basis of evidence, how oversight adapts to risk and how accountability is maintained, and those issues should be addressed before agentic AI becomes more deeply integrated into clinical practice.
Why it matters
The framework arrives as adoption is running ahead of governance. A survey of 250 U.S. healthcare leaders cited by Healthcare Dive found that about 28 percent said their organization had already deployed agentic AI tools and about 44 percent were running pilots or proof-of-concept projects. Seventy-two percent said such tools are occasionally implemented without IT approval, and more than half named compliance or regulatory violations and unauthorized access to protected health information as major concerns.
There is no substantive federal regulation specific to health AI, and states have produced a patchwork of laws, Healthcare Dive reported, citing the law and consulting firm Manatt’s count of more than 250 AI bills affecting health care introduced across 47 states in 2025. In that gap, professional bodies have been issuing their own guardrails. For physicians, the AMA’s framing matters because it ties the question of what an AI may do to the question of who answers when it fails, which bears directly on malpractice liability and on licensure, the two areas the AMA says it will work on next.
What to watch
The AMA says the four questions will guide its engagement with developers, regulators and policymakers on liability, licensure and regulatory approaches, but it has not set a timetable for specific policy proposals. The framework follows two earlier AMA interventions this year, as reported by Digital Health News: an April call for lawmakers to strengthen safeguards on AI chatbots used in mental health services, and a May policy framework against unauthorized AI deepfakes and medical impersonation.
The caveats
The framework is a position document from a physician membership organization, not a regulation or a validated technical standard. The Whyte comments are reported second-hand by Digital Health News from a statement he gave to Fierce Healthcare, and are paraphrased here rather than quoted. The adoption figures come from a survey of healthcare leaders, not physicians, cited by Healthcare Dive, which did not name the survey’s sponsor.
This article is for general information and does not replace advice from a qualified clinician. Corrections: email the editor using the address on the Contact page.
