Inside the state regulatory sandbox that waived normal prescribing rules and put oversight in the hands of a five member AI board with no physicians.
Utah's own medical licensing board says it learned about a state-run AI prescription refill pilot from news coverage, and eleven of its members wrote to state officials in March asking for the program to be halted. The pilot, which lets Utah residents renew certain prescriptions through an AI chatbot instead of an in-person doctor visit, is the first publicly disclosed state program in the country to put a large language model in the prescribing seat.
The pilot runs inside a Utah "regulatory sandbox" that waived normal prescribing rules for participating companies. A five-member AI-specialist board, set up specifically to oversee sandbox experiments and containing no physicians, approved Doctronic's entry in December 2025. The state announced the partnership in January, billing it as a way to expand access to prescription renewals without forcing patients to schedule a clinic visit for routine refills.
These are prescription refills, not lifestyle tips. The drugs involved carry side effects and drug-interaction risks, and the company's roadmap moves the pipeline from doctor-reviewed refills to fully automated ones. The next decision point is whether Utah will allow the program to scale beyond its initial scope.
Utah and Doctronic launched the pilot earlier this year under a Department of Commerce partnership. The mechanism is a state regulatory sandbox, a legal carve-out that lets companies test products without meeting the licensing requirements that would normally apply. For medicine, that means a product can operate without holding the century-old gate that limits prescribing to licensed physicians. The oversight body, populated by AI specialists and ethicists, has no clinicians with prescribing experience.
Doctronic is a venture-backed consumer AI health product that markets itself as a "24/7 digital doctor" (company page; LSVP investor profile). Its public materials say the chatbot screens patients, identifies candidates for renewal, and either recommends or issues a refill. The pilot's original design included a doctor review step; the company's stated plan is to remove that step as the program expands (Ars Technica).
Doctors' objections have hardened since the launch. The Utah Medical Licensing Board, the body that polices prescribing in the state, sent a formal letter to state officials around April 2026, distributed by the Federation of State Medical Boards (FSMB letter PDF), arguing that any AI system allowed to prescribe must meet standards "akin to human doctors" and that the sandbox structure removed the normal accountability layer.
Dr. Eric Bressman, a physician who has written about AI in medicine, put the concern in concrete terms: the program has "crossed a threshold in terms of giving something that is not human a medical license" (Boston Globe; KUER). The doctors' core argument is procedural as much as clinical. Accountability for a bad outcome has no clean answer when the prescriber is a model. Who is named on the prescription? Who is sued when something goes wrong?
The accountability gap is structurally built in. Sandbox rules let a company operate without holding the medical license that would normally anchor malpractice, supervision, and continuing-education duties. Doctronic publishes a description of its safety process but does not publish peer-reviewed clinical validation data. The Utah Department of Commerce called the program "groundbreaking." State law does not require a sandbox participant to disclose error rates, scope of conditions covered, or escalation triggers.
The state's own licensing board has not received a formal response to its March letter, according to coverage from the Boston Globe and KUER. The next decision point is not theoretical: the company has said it intends to move the Utah program from doctor-reviewed to fully autonomous refills, and the sandbox board is the body that would approve that change.
Independent clinical evidence on autonomous AI prescribing is thin. Most published evaluations of AI diagnostic and prescribing tools test clinician-assist settings, where a doctor reviews the model's output before acting. Utah's program is testing what happens when the model is the prescriber.