Medicine already lets a supervised non-physician inside the confidential encounter. Utah has twice let a supervised AI act in care, by signed agreement.
We are preparing the next ask.

What is being prepared, precisely.

First, a sandbox agreement: a regulatory mitigation agreement under the AI Policy Act authorizing a physician-supervised care agent for an enumerated list of delegated acts — intake, prescription-renewal preparation, chronic-care follow-up, explanation of results — in an access-to-care setting, with the same phased supervision taper the renewals agreement already uses, published metrics, and a written trigger for conversion to a durable rule, because a mitigation agreement is a twelve-month instrument and a precedent should outlive its paperwork.

Second, licensing guidance: recognition from the Division of Occupational and Professional Licensing that a certified, registered, physician-supervised agent — disclosed to and accepted by the patient — is a permissible delegate under the practice act, with the physician carrying responsibility. That is not our frame; it is the Federation of State Medical Boards' own 2024 policy: boards regulate physicians, not tools.

Third, an evidence-rule confirmation: a committee note or one-clause amendment, through the Utah Supreme Court's public rules process, confirming that participation of an authorized, physician-directed agent falls within Rule 506's existing clause and does not, standing alone, waive the privilege.

The trust artifact underneath all three is the same: a tamper-evident audit trail — which physician directed which task for which patient, when — held as medical-record material, readable by a regulator without interviewing anyone.

This shape is a deliberate answer to how Utah's first renewals agreement drew fire. The board there did not object to a machine touching care; it objected to two things — being told after the system was already live, and a refill going out without the reassessment a physician owes the patient. This ask is built to fail neither test. The order is inverted: the profession asks first, through its own bodies and on its own record, before anything is filed — the board's stated wish that "professionals with medical backgrounds review all proposals prior to implementation," made the starting move rather than the grievance. The physician never leaves the loop, because every delegated act runs under a named physician's direction and the acts that define the profession stay in human hands. And the audit trail lets the board confirm both without taking anyone's word for it. It is the board's own invitation, taken literally: to safely implement AI in the practice of medicine.

None of these instruments is ours to file. Sandbox agreements go to operators with a named supervising physician; delegation guidance is the profession's to request; the evidence rule changes through the court's own public process. The Foundation prepares the research, the drafting, and the audit infrastructure — and offers them to the physicians who lead. The profession leads because, in Utah, the record already shows what happens when it does not.

Utah needs the access more than most.

By the AAMC's workforce data it sits in the bottom tier of states for physicians per capita, and published federal projections give it the worst obstetric-gynecology workforce adequacy in the nation by 2037 — under half of projected need. Nationally, 92.3 million people live in primary-care shortage areas and 137.1 million in mental-health shortage areas.

The design is the one medicine already trusts, with one addition. Roughly 477 student-run free clinics operate in this country: students see patients the market does not serve, and a supervising physician reviews and signs. In the pilot, the supervised agent joins that room — preparing intake, renewal paperwork, and follow-up under the named physician's direction, on the same phased review the state has already used. The patient is told, and accepts. The acts that define the profession never move: diagnosis, the prescribing decision, and escalation stay with the physician — the same lines Utah's own agreements drew. No training on patient data and no sale of it, the confidentiality floor Utah itself wrote into its 2025 mental-health-chatbot statute. Every step logged. Every metric published. Care free at the point of delivery.

The pilot's product is the record: access delivered and hours returned, in numbers a department and a court can read before anyone asks either to make anything permanent.

In law, the precedent had to be built.

Minnesota's supreme court put supervised law students inside the attorney-client privilege in 2013, then legal paraprofessionals in 2020 — as a pilot that measured itself, reported, and became permanent. The third ask being prepared there is the lawyer's supervised AI agent, on the same four conditions every prior rung used: certified, supervised by a named professional who carries personal responsibility, disclosed to the client and accepted, bound to confidentiality. Those are also, almost word for word, the conditions the FSMB's policy and Utah's signed agreements already impose in medicine.

The law campaign also studied why Utah's own legal sandbox — first in the nation — is winding down, and the answer was not harm: the published complaint rate was vanishingly low. It was design: bespoke hand-vetting that drowned a small office, entrants using authorization as marketing, permanence resting on champions. The lesson is registration and machine-readable reporting over bespoke vetting, no endorsement claims, and permanence tied to published metrics. Utah's AI office already runs on those mechanics — registration, required reporting, published agreements — which is part of why medicine's ask lands here first.

In April 2024 the Federation of State Medical Boards adopted policy guidance on incorporating AI into clinical practice — physician accountability, disclosure, consent — and stood up a workgroup on AI regulation.

The AMA's surveys show physician use of AI rising from 38 percent in 2023 to 66 percent in 2024 to more than 80 percent by 2026; the profession's own term for it is "augmented intelligence." Texas has enacted a statute expressly authorizing practitioner use of AI in diagnosis with records review and patient disclosure. And even the states that banned AI-delivered therapy in 2025 preserved the supervised, professional-directed lane — the line this program stands on, codified by the most cautious legislatures in the country.

The list, in one paragraph.

Utah's two clinical mitigation agreements and its pilot pages, from the Office of Artificial Intelligence Policy at commerce.utah.gov, including the January 2026 renewal agreement read in full. The Utah Medical Licensing Board's April 2026 letter calling for that program's suspension — read in full, and the source of every phrase quoted from it above — and the state's response declining to suspend, both from commerce.utah.gov. Utah Rule of Evidence 506, from the state courts' published rules. The AI Policy Act (SB 149, 2024) and the mental-health-chatbot statute (HB 452, 2025), from le.utah.gov. The FSMB's 2024 AI policy and workgroup announcement, from fsmb.org. The AMA adoption surveys, from ama-assn.org. The shortage-area populations, from HRSA's quarterly reports. The physicians-per-capita standing, from the AAMC's workforce dashboard, and the 2037 OB/GYN projection from the published federal workforce study. The ambient-AI scale figures, from NEJM Catalyst. HIPAA's treatment and business-associate provisions, from the eCFR. Where this page leans on the law campaign's Minnesota and Utah-sandbox record, those sources are set out at law.org.ai/mn.