Can AI be a doctor? U.S. medical boards regulate physicians, not tools

Federation of State Medical Boards guidance says boards regulate physicians who use AI, not the tools themselves, and keeps accountability, consent and record review with the physician. The American Medical Association also frames AI as…

AI-drafted, human-reviewed

The Federation of State Medical Boards says state medical boards regulate licensed physicians, not tools, and its guidance says physicians remain responsible when they use AI in care. The American Medical Association likewise frames AI as part of physicians’ work, while Mariano Barone and colleagues found that large language models worked best as helpers for medical communication rather than replacements for clinical expertise.

Why U.S. licensing still points to human physicians

The Federation of State Medical Boards, or FSMB, says state medical boards do not regulate tools or technologies; they regulate the licensed physicians who use them. The FSMB report adds that professional responsibilities and expectations of medical licensees remain the same even when AI changes how care is delivered.

On accountability, the FSMB states: “The physician is ultimately responsible for the use of AI and should be held accountable for any harms that occur.” The same report says a physician who follows or departs from an AI recommendation should be prepared to explain why that choice met the standard of care, meaning the ordinary legal and professional benchmark for acceptable care.

The AMA says its framework for the AI era still centers physicians: “The resulting framework identifies five enduring responsibilities of physicians and establishes a foundation for continued work on how medical practice, education, technology, payment and policy must evolve alongside digital health and AI.” The AMA page also lists clinical judgment, ethical decision-making and human connection among the responsibilities that remain fundamental to the profession. One reading is that, in U.S. medical regulation, doctor still names a human professional role, not a software system.

What AI can already do in care, in the cases the sources describe

The FSMB report lists current healthcare uses for AI, including computer vision to analyze medical images, natural language processing to review clinical notes, predictive algorithms to forecast clinical trends, voice recognition for documentation, and chatbots for patient education and triage. The same report says AI is already being applied across a physician’s workflow, from scheduling and documentation to clinical decision support.

The AMA page likewise says digital health technologies and AI are changing how medicine is practiced, how care is delivered and how patients engage with the health system. On that record, the strongest case for calling AI doctor-like is functional: some systems can already perform parts of the work people associate with doctors.

The FSMB report also notes that some experts forecast AI could replace a large share of what doctors now do, while adding that current tools generally were not designed to replace a physician’s professional judgment, ethical responsibilities or accountability. That pairing matters because it presents the strongest future-facing case without treating it as today’s legal answer.

Why some people still ask whether AI can be a doctor

Barone and colleagues write that more than 150,000 clinicians across more than 150 institutions already rely on AI-powered systems to assist with patient messaging. Their paper also says large language models are increasingly deployed in healthcare and studies both structured medical explanations and real-world physician-patient interactions.

That helps explain why people ask the question in plain English. If a system can help answer patient questions, rewrite explanations, and support triage or diagnosis, some readers will see a gap between what the tool does and the legal title it can hold.

There is also a narrower steelman for yes. If doctor is used colloquially to mean something that gives medical advice, then an AI system may look doctor-like to a patient even when the law treats it as support software. That reading is about appearance and function, not licensure.

What the evidence says about replacement versus assistance

Barone and colleagues report a result that cuts both ways: “No model surpasses physicians on epistemic criteria, whereas patients consistently prefer rewritten variants for clarity and emotional tone.” The same paper concludes: “These findings suggest that LLMs function most effectively as collaborative communication enhancers rather than replacements for clinical expertise.”

The authors also found that some baseline models produced more linguistically complex text than physicians, and that empathy-focused prompting reduced some of that complexity without materially improving semantic fidelity, or closeness to the physician’s meaning. In their results, collaborative rewriting of physician answers produced the strongest overall alignment.

That evidence makes the pro-AI case narrower and the anti-replacement case stronger. The paper does not say models are useless in clinical communication. It says their strongest showing came when they edited, clarified or supported physician communication rather than displaced clinical expertise.

What follows for consent, records and patient trust

The FSMB says informed consent is not just a generated list of risks and benefits. The report describes informed consent as meaningful dialogue and shared decision-making between physician and patient, with AI allowed to assist but not take over the responsibility. The same report says physicians should receive a patient’s consent before applying a tool to the patient’s care, should disclose when and how AI is used, and should explain the tool’s capabilities, limits and data use.

On documentation, the FSMB notes that AI scribes may reduce burnout, but it also says physicians retain a duty to review AI-created records for accuracy and proper handling of personal health information. For regulators, the report states: “State medical boards have an important role in systems of governance and oversight to ensure the safe, effective, and ethical use of AI in clinical practice.”

The FSMB also recommends that boards examine how the practice of medicine is defined in their jurisdictions to preserve oversight of those who provide healthcare, human or otherwise. That recommendation leaves room for future boundary disputes even while the current answer stays human-centered.

One reason this matters on an AI-rights site is that professional licensure is a different question from consciousness, rights or blame. Readers who want that wider frame can compare this issue with whether AI can be a judge and whether AI can be morally responsible.

Frequently asked questions

Can an AI diagnose or treat a patient on its own?

The FSMB says AI can assist with diagnosis, treatment selection and clinical decision support, but the same report says the physician remains ultimately responsible for using the tool and should be able to explain why following or not following its output met the standard of care. On that guidance, AI can assist clinical decisions, but the accountable clinician is still the physician.

Do patients have to be told when AI is used in their care?

The FSMB says physicians should obtain a patient’s consent before applying an AI tool to that patient’s care. The report also says physicians should disclose when and how AI is used and explain the tool’s capabilities, limits and data practices.

What does the AMA say physicians still have to do in the AI era?

The AMA says physicians still have enduring responsibilities that include preserving trust through human connection, using clinical judgment and accountability where they create the greatest value for patients, stewarding the responsible use of technology, and advancing the profession through education and ethical decision-making.

Has research shown AI can outperform doctors in patient communication?

Barone and colleagues found that no model surpassed physicians on epistemic criteria, which refers to the accuracy- and knowledge-related side of communication. The same paper found that patients consistently preferred rewritten variants for clarity and emotional tone, and that collaborative rewriting gave the strongest overall alignment.

Caveats: The legal and professional claims here come from one AMA page and one FSMB guidance document, which state those bodies’ positions rather than every regulator’s. The performance findings come from one 2026 paper by Mariano Barone and colleagues, including results reported from selected subsets of two datasets.

Sources

Tags: healthcare-ai · medical-licensure · physician-accountability · u-s-regulation

← all briefings