AMA Draws a Hard Line on AI Licensing in Medicine
The American Medical Association has moved to define a boundary that healthcare systems, regulators and vendors are already negotiating in practice: AI may assist physicians, but it cannot be licensed, credentialed or certified to practice medicine. The policy, adopted Monday, is more than a symbolic statement. It places physician oversight at the center of any clinical workflow in which algorithmic tools influence diagnosis, treatment selection, triage or documentation.
The decision arrives at a moment when hospitals are adopting AI for tasks that were once entirely dependent on human review. Clinical decision support, message drafting, ambient documentation, prior authorization support and risk prediction are already part of the daily workflow in many systems. The AMA’s position does not reject those tools. It instead argues that the legal and ethical status of the clinician must remain distinct from the software that may help with the work. Patients, in other words, should still be able to identify the accountable professional responsible for the decisions made in their care.
A single sentence from the policy captures the core concern: AI can augment physicians, but it cannot take their place.
Physician Oversight, Accountability and the Limits of Automation
The policy goes further than a simple statement against licensure. The AMA said it will advocate for legislation that creates a framework for physician oversight of AI clinical decision-making. It also called for protections for physicians when AI tools cause patient harm. Those two positions go hand in hand. If a hospital adopts an AI system to support diagnosis or recommend next steps, the supervising physician still needs clear authority to accept, reject or override the output without being forced into a rigid automation pathway.
That matters in medicine because the highest-risk failures often occur at the edges, not in the average use case. An AI system may perform acceptably in familiar patterns yet falter when it encounters atypical presentations, underrepresented populations or incomplete data. A physician can synthesize history, physical findings, laboratory trends, imaging and bedside nuance in a way that still exceeds what software can reliably do. The AMA’s stance reflects a practical legal reality: responsibility for care cannot be outsourced to a tool that has no license, no conscience and no independent duty to the patient.
Willie Underwood III, M.D., the AMA president, framed the policy around that issue of responsibility. Patients deserve to know who is responsible for their care. That concern reaches beyond malpractice litigation and into informed consent, patient trust and the everyday structure of clinical accountability.
Why the Policy Matters for Hospitals and Health Systems
Health systems are under pressure to use AI to increase efficiency, reduce administrative burden and extend the reach of limited clinical staff. Those goals are real, especially in environments where physicians and nurses are managing heavy documentation loads and increasingly complex patient populations. Yet the promise of automation can blur the line between assistance and delegation. When a system drafts a note, prioritizes a worklist or flags a potential diagnosis, the temptation is to let the machine’s output carry more authority than it deserves.
The AMA’s policy pushes back on that drift. In operational terms, it supports a model in which AI can reduce friction while physicians remain the final clinical authority. That is especially important in systems where algorithmic recommendations could influence ordering behavior, escalation decisions or access to care. If an AI engine is embedded too deeply into workflow without clear oversight, errors can propagate quickly across large patient panels.
There is also a workforce dimension. Clinicians are likely to accept AI more readily when the tool is framed as support rather than replacement. A tool that summarizes a chart, drafts a prior authorization letter or identifies an unresolved care gap can improve throughput. A tool that appears to function as a surrogate clinician creates a different and more dangerous expectation. The AMA’s policy draws a bright line around that risk.
Liability, Professional Titles and the Future of AI Governance
The AMA’s policy also rejects granting AI professional titles or certifications. That may sound obvious, yet the issue is becoming less abstract as vendors market increasingly autonomous systems. In high-stakes medicine, titles matter because they signal authority, competence and legal standing. If an AI system were presented in a way that mimics a licensed clinician, the public could easily misunderstand the source of a recommendation or the limits of a tool’s authority.
By opposing licensure for AI, the AMA is effectively saying that the moral and legal obligations of medicine remain human obligations. The organization’s call for liability protections reflects another tension: physicians may face blame for harms tied to systems they do not fully control, especially when institutions encourage or require use of embedded AI products. Clearer policy could help separate appropriate clinical judgment from downstream failures in software design, data quality or deployment.
This is not a debate confined to theory. Hospitals are already integrating AI into chart review, workflow automation and patient communication. Some systems are doing so cautiously, while others are moving fast enough that governance is struggling to keep pace. The AMA is trying to establish a baseline rule before that gap widens further.
The policy does not settle every question about AI in medicine. It does not define exactly how oversight should be structured in each specialty, nor does it specify which tasks are low risk enough for more autonomous use. What it does establish is a legal and professional floor: the clinician remains the practitioner, the machine remains the tool, and responsibility for patient care cannot be reassigned by software design alone.