92.3 million Americans live where there are not enough primary-care physicians to go around.
We are writing down what a physician's supervised AI may now do about that — and what stays with the licensed human.

Medical students take histories and write notes; Medicare has billed on student documentation, with physician verification, since 2018.

Physician assistants practice on delegated authority. Human scribes type in the corner. And since 2023, ambient AI scribes have listened and drafted the note in millions of visits under business-associate agreements — roughly 2.5 million encounters in one California medical group alone — with no confidentiality crisis following, because the law already shelters those who help under the physician's direction.

Every rung of that ladder is a non-physician admitted to the room on the same conditions: the physician directs, the physician reviews, the physician answers for it.

The protection is real but layered.

Federal law contains no general physician-patient privilege at all — the Supreme Court has recognized only the psychotherapist-patient privilege (Jaffee v. Redmond, 1996) — so the evidentiary shield lives state by state. Meanwhile HIPAA already permits disclosures for treatment and already admits technology vendors as business associates under contract; that is how the listening AI got into millions of encounters lawfully.

And in a family of states, the privilege rule itself already anticipates the question. Utah's Rule of Evidence 506, and its counterparts in Texas, New Mexico, and North Dakota, extend the confidence to persons "participating in the diagnosis or treatment under the direction of the physician." Medicine wrote its agents clause decades ago. The words are already there.

The agent is the physician's supervised instrument in medical care — never a caregiver, never a therapist.

Before it touches anything about your care, a licensed physician is engaged: yours, named, carrying personal professional responsibility, which is exactly where the Federation of State Medical Boards' 2024 policy puts accountability for AI in clinical practice. You are told the agent participates, and you accept. The acts that define the profession stay with the physician: diagnosis, the prescribing decision, and escalation the moment anything is not routine.

Every direction is logged — which physician directed which task for which patient, when — in a tamper-evident audit trail held as medical-record material, because that record is what a regulator, or a court, would read. And the system does not learn from your care: no training on patient data, no sale of it — not as a policy promise but as an architecture that can be inspected.

Under its 2024 Artificial Intelligence Policy Act, Utah's Office of Artificial Intelligence Policy signs regulatory mitigation agreements — published contracts that admit a specific AI deployment into regulated practice under specific conditions.

In January 2026 it signed one authorizing AI-prepared prescription renewals for roughly two hundred non-controlled medications, with physician oversight phased from full pre-issuance review toward audited samples — co-signed by the state's physician licensing division. Under another, an AI prepares psychiatric-medication renewals with patient consent, audits, and escalation triggers. Utah's evidence rule already reaches persons participating in treatment under the physician's direction. And the office that signs the agreements and the division that licenses physicians sit in the same Department of Commerce.

law.org.ai is the Foundation's program for legal help: a published catalog of the acts reserved to licensed lawyers, a supervision design, and a first state — Minnesota, whose supreme court has twice let a certified non-lawyer inside the attorney-client privilege, students in 2013 and paraprofessionals in 2020, and where the third ask is being prepared.

med.org.ai runs the same design in medicine: reserved acts with the physician, a supervised agent, a disclosed and consenting patient, an audit trail, a pilot that measures itself in public. Medicine's ladder is, if anything, taller — its privilege rules already name persons acting under the physician's direction, and Utah has already signed.

The list, in one paragraph.

The shortage-area populations, from HRSA's quarterly designation reports. The wait-time average, from AMN Healthcare's 2025 survey of physician appointment wait times. The two-hours-of-record-work finding, from the 2016 Annals of Internal Medicine time-and-motion study (Sinsky et al.). The ambient-AI scale figures, from NEJM Catalyst's report on the Permanente Medical Group. Medicare's student-documentation rule, from the 2018 CMS change. Utah's mitigation agreements, from the Office of Artificial Intelligence Policy's published agreement and pilot pages at commerce.utah.gov, including the January 2026 renewal agreement in full. Utah Rule of Evidence 506, from the state courts' published rules. The AI Policy Act (SB 149, 2024) and the 2025 mental-health-chatbot statute (HB 452), from le.utah.gov. HIPAA's treatment-disclosure and business-associate provisions, from the eCFR. The FSMB's 2024 AI policy, from fsmb.org. Jaffee v. Redmond, from the published opinion.