State medical boards move first as autonomous clinical AI enters the room
Behind closed doors in Washington, a 22-member workgroup is trying to define how medicine should govern AI systems that can act with limited or no direct physician supervision. The effort sits at the intersection of licensure, patient safety, and corporate influence, with the Federation of State Medical Boards positioning itself as the body most likely to turn a policy draft into state-level standards.
The roster alone explains the stakes. Executives from the American Medical Association, the American Hospital Association, AI company Anthropic, and insurer Elevance Health are all part of the group, alongside representatives from state governments and other sectors tied to medical regulation, AI development, and care delivery. In practical terms, the people helping write the rules include those likely to be regulated by them.
One sentence captures the pressure points. The workgroup is not debating whether AI belongs in medicine; it is debating how much autonomy it should have before state boards treat it as part of the practice of medicine.
Limited supervision is the regulatory fault line
The stated goal is to draft guidance for local medical boards, with particular attention to AI tools that perform clinical functions with limited or no direct physician supervision. That phrase matters because it marks the boundary between software that assists clinicians and software that begins to behave like a clinical actor. Once an algorithm is recommending diagnoses, triaging symptoms, or shaping treatment decisions without constant human review, the legal and ethical questions shift from procurement to licensure.
Humayun Chaudhry, president and chief executive of the FSMB, said the recommendations are being developed to help state boards regulate that emerging class of tools. A draft could surface in weeks to months. If adopted, the recommendations could become a template for state legislation, extending the influence of the workgroup far beyond Washington conference rooms.
State medical boards regulate medical licensure and the practice of medicine individually, which makes this a fragmented policy arena. A unified national standard does not exist, and that vacuum gives the FSMB unusual leverage. In a field where software iterations move faster than legislatures, even nonbinding guidance can become the de facto rulebook if enough states copy it.
Confidential talks, public consequences
The discussions inside the workgroup are confidential. That is not unusual for a policy process still at the drafting stage, but secrecy carries a different weight when the subject is autonomous clinical AI. These systems are already entering workflows across care settings, and their regulatory status remains unsettled. Without clear state guidance, boards may struggle to distinguish experimental deployment from unlicensed practice, especially when tools are embedded in clinical pathways rather than sold as standalone devices.
The power struggle is easy to see. Physician groups want guardrails that preserve accountability and keep AI from substituting for clinical judgment. Hospitals want usable standards that do not make deployment impossible. Tech companies want rules flexible enough to accommodate rapid product changes. Insurers have their own incentive to support tools that promise efficiency while limiting downstream liability. State governments want a framework they can enforce without building a new regulatory architecture from scratch.
Autonomy is the central issue because medicine has always required a named professional to own the decision. If an AI system can act with little supervision, the old chain of responsibility becomes harder to trace. When an adverse event occurs, the question is no longer only whether the clinician followed the software’s recommendation. It becomes whether the software was functioning inside a lawful scope at all.
Why the AMA and others want a seat at the table
The American Medical Association’s presence signals how seriously organized medicine views the regulatory battle. The AMA has long argued that clinical decision-making should remain anchored in physician oversight, and its inclusion in the workgroup gives doctors a direct voice in defining where AI support ends and clinical substitution begins. The American Hospital Association brings a different perspective, shaped by operational realities, staffing pressure, and the need to integrate new tools into hospital systems without destabilizing care delivery.
Anthropic’s role is equally revealing. AI developers want a say before states impose rigid rules that could slow product rollout or force expensive redesigns. Elevance Health, representing a major insurer, suggests that payment and utilization management concerns are also part of the conversation, even if they are not explicit in the draft language. When an insurer sits beside a medical association and a model developer, the question is not only who may use AI, but who may pay for it and under what safeguards.
The workgroup’s composition also reflects a broader shift in health tech regulation. Oversight is no longer confined to federal agencies. State boards, often overlooked in national debates, may end up deciding how far autonomous clinical AI can go at the bedside, in telehealth encounters, and in other settings where a patient may never know how much of the encounter was machine-driven.
A draft in weeks could set off state-by-state adoption
A public draft in the coming weeks or months would likely trigger a cascade of responses from medical societies, hospital systems, vendors, and lawmakers. Because state boards influence licensure and the practice of medicine, even recommendations without the force of law can shape enforcement priorities, inform disciplinary actions, and influence future statutes. In a regulatory environment this unsettled, soft law often becomes hard reality.
The most consequential detail is not the closed-door setting. It is the scope of the task. The workgroup is trying to decide how to regulate tools that may already be functioning as quasi-clinicians, with limited physician supervision and growing clinical reach. That makes the next draft more than a policy memo. It could become the first widely used blueprint for how states treat autonomous clinical AI inside the practice of medicine.
What remains unresolved is how much oversight states will demand before software is allowed to meaningfully shape care on its own, and how boards will enforce those rules once the tools are already embedded in clinical workflows.