"Artificial intelligence should not be licensed to practice medicine independently; instead, generative AI should be governed within existing professional accountability structures rather than treated as a separate practitioner."
This pivotal statement, issued by the leaders of the organization representing the nation’s state medical boards, directly addresses a growing query from state legislatures regarding the integration of artificial intelligence into healthcare. The Federation of State Medical Boards (FSMB) has articulated a clear stance: AI tools, particularly generative AI, should not be afforded the same independent licensure as human physicians. Instead, the FSMB advocates for their governance and oversight through established professional accountability mechanisms, emphasizing that human clinicians must remain ultimately responsible for patient care and any harm arising from AI’s use. This position carries significant weight, shaping the ongoing discourse on how these rapidly evolving technologies can be safely and ethically incorporated into the practice of medicine.
The Federation of State Medical Boards (FSMB), through its president and chief executive, Humayun J. Chaudhry, and its board chair, Christy Valentine Theard, has published a compelling argument against the independent licensure of artificial intelligence in medical practice. Their recent commentary, appearing in STAT, underscores a fundamental distinction between licensing human practitioners and authorizing AI tools. The FSMB posits that medical licensure is more than a simple certification of competence; it is rooted in a profound social contract between society and physicians. This contract bestows upon physicians unique privileges, such as the authority to diagnose, treat, and prescribe, in exchange for stringent duties of competence, ethics, and accountability. A license, therefore, represents a commitment to these elevated standards and responsibilities, a framework that the FSMB argues cannot be directly applied to AI systems.
The core of the FSMB’s argument rests on the principle of accountability. They propose that responsibility for AI’s actions should be commensurate with its level of autonomy. While AI can serve as a valuable tool, generating insights or drafting documentation, the ultimate responsibility for patient outcomes must remain with the licensed human clinician. This means that physicians should be held accountable for any harm that occurs due to their inappropriate reliance on AI, even when the technology itself performs as designed. This framework aligns with existing nonbinding policy guidance adopted by the FSMB House of Delegates in 2024, which emphasizes the responsible and ethical incorporation of AI into clinical practice. This guidance encourages continuing education for healthcare professionals on AI, reinforces the enduring professional responsibility of the licensee, and addresses accountability for AI use across various clinical settings.
The timing of the FSMB’s statement is not coincidental; it is a direct response to burgeoning legislative activity at the state level. This year, both Idaho and Iowa introduced bills that explored the creation of separate licensing structures for artificial intelligence-augmented and autonomous service providers, distinct from the existing medical board system. Idaho’s House Bill 945, for instance, proposed a new Board of Autonomous Medical Practice with its own licensure program and a regulatory sandbox. Iowa’s House Study Bill 766 outlined a comprehensive licensing and oversight framework for similar entities. While neither bill has yet been enacted, their introduction signifies a tangible shift towards considering AI as a potentially independent medical actor, prompting the FSMB to articulate its position clearly.
Utah has taken a more concrete step with its pilot program. In January, the state’s Office of Artificial Intelligence Policy announced a 12-month agreement with the health technology startup Doctronic. This initiative allows an autonomous AI platform to participate in prescription renewals for patients with chronic conditions like hypertension, diabetes, and thyroid disease, operating within the state’s regulatory sandbox. Crucially, this arrangement removes the requirement for a clinician to sign off on each individual renewal, with uncertain cases being escalated to human clinicians. However, this pilot has not been without its detractors. The Utah Medical Licensing Board expressed significant concerns, calling for the program’s suspension due to potential patient safety risks. The FSMB leaders highlight this reaction as a critical illustration of why medical boards must have a voice in situations where clinical decisions and patient safety are involved, even when innovations are processed through state technology offices rather than traditional medical regulatory channels.
In response to these evolving challenges, the FSMB has established a Workgroup on the Regulation of AI in the Practice of Medicine. This group is tasked with updating the Federation’s guidance to reflect the current wave of AI systems that are pushing towards greater autonomy. The existing framework, while foundational, predates the widespread development of these more autonomous AI capabilities, necessitating a review and potential revision.
The FSMB’s stance is not the only perspective in this complex debate. A significant counterargument, published in peer-reviewed journals, advocates for a more direct approach to AI licensure. In JAMA Internal Medicine, researchers led by Eric Bressman of the University of Pennsylvania, including Carmel Shachar, Ariel D. Stern, and Ateev Mehrotra, proposed licensing AI in a manner analogous to clinicians. They draw parallels between current concerns about AI’s potential for hallucination and performance drift and historical anxieties regarding variable clinician training that licensure was originally designed to address. Their proposed framework includes practice standards, ongoing surveillance, and continuing education, ideally overseen by a new federal digital licensing board, with the FDA retaining premarket review responsibilities to avoid a fragmented state-by-state regulatory landscape.
A separate viewpoint in JAMA, authored by Alon Bergman, Robert Wachter, and Ezekiel Emanuel, presented a four-part licensure framework specifically for autonomous clinical AI. This framework includes demonstrated competency, requiring AI models to meet or exceed the performance of recent human test-takers on licensing examinations. It also mandates a supervised deployment phase, akin to medical residency, a clearly defined scope of practice, ongoing monitoring, and periodic renewal. Their argument is partly fueled by projected physician shortages that traditional workforce solutions may struggle to address in a timely manner, suggesting AI could fill critical gaps.
Both camps acknowledge the inherent risks associated with AI in healthcare. The fundamental disagreement lies in whether the safer and more effective path forward involves adapting traditional licensure models for AI or extending existing accountability structures to encompass AI use.
For patients, the immediate implications of this debate, while currently conceptual, are practical. Firstly, under existing state frameworks, any AI system involved in a patient’s care still places legal accountability squarely on the shoulders of the licensed clinician. This status quo, defended by the FSMB, ensures that patients have a clear avenue for recourse should errors occur. Secondly, patients have the right to inquire about the involvement of AI in their care. FSMB guidance encourages physicians to be transparent about AI usage, making questions about AI’s contribution to diagnoses, clinical notes, triage decisions, or prescription renewals entirely reasonable.
Thirdly, the degree of autonomy exercised by an AI system is a critical distinction. A tool that drafts a clinical note for a physician’s review represents a different regulatory challenge than a system that autonomously renews prescriptions without individual clinician oversight. The Utah pilot program, which leans towards the latter, is precisely why it has drawn the attention of medical boards concerned about patient safety.
The ongoing debate is poised to evolve significantly as state legislatures reconvene. The proposed legislation in Idaho and Iowa is unlikely to be isolated, with similar initiatives expected. The FSMB workgroup’s updated guidance will undoubtedly influence how individual state medical boards respond, though it is important to remember that this guidance is advisory, and each board retains its statutory authority. The 2024 FSMB policy urged boards to critically examine how the practice of medicine is defined within their jurisdictions—a question that becomes increasingly consequential as AI systems operate with less direct human supervision. This definitional work is the mechanism through which this complex debate will ultimately translate into concrete regulations.
For the public and healthcare professionals alike, the most instructive aspect of this discourse will be the observation of specific pilot programs. Areas like prescription renewal, patient triage, and diagnostic suggestion are at the forefront of AI integration. These are precisely the tasks where even a routine decision, if mishandled by an AI or through over-reliance by a human, can lead to tangible harm. The outcomes of these early initiatives will provide critical data and insights that will inform future policy and regulatory decisions, shaping the responsible integration of AI into the fabric of healthcare delivery.
Frequently Asked Questions
What did the Federation of State Medical Boards say?
The FSMB’s president and board chair stated that AI is not yet ready for independent licensure akin to physicians. They advocate for AI to be governed within existing professional and institutional accountability structures.
Does FSMB have the power to decide this?
No, the FSMB is an advisory body that serves the state and territorial medical boards. Each individual board retains its own statutory licensing authority, and state legislatures are responsible for enacting the underlying laws.
Has any state tried to license AI as a practitioner?
Bills introduced in Idaho and Iowa this year explored the possibility of establishing separate licensing structures for AI-augmented and autonomous service providers. Neither of these legislative efforts has been enacted into law.
What is the Utah pilot?
Utah’s Office of Artificial Intelligence Policy has approved a 12-month program in partnership with Doctronic. This initiative allows an autonomous AI platform to manage certain prescription renewals for patients with chronic conditions, without requiring a clinician to approve each individual request.
Who disagrees with the FSMB position?
Researchers publishing in JAMA Internal Medicine and JAMA have proposed adapting licensure frameworks for autonomous clinical AI. Their proposals include requirements such as competency testing, a supervised deployment phase analogous to residency, and a defined scope of practice.
Who is responsible if AI contributes to a medical error today?
Under current state regulations, the licensed clinician remains legally accountable for patient care, even when AI is involved. FSMB guidance emphasizes physician responsibility for the use of AI in clinical practice.
Can I ask whether AI was used in my care?
Yes, patients are encouraged to ask. Inquiring about whether an AI tool contributed to a diagnosis, clinical note, or prescription decision is reasonable. FSMB guidance promotes transparency regarding the use of AI in patient care.
