A 'regulatory sandbox' lets the state waive rules for AI companies. Eleven board members urge a halt, and a think tank is pitching the model to other states.
Utah didn't wait for its medical board to decide whether AI could prescribe drugs. It built a legal door called a regulatory sandbox, opened the door, and walked through it first.
A regulatory sandbox is a state-created zone where officials can temporarily waive specific laws so a company can test a new product under supervision. Utah's Office of AI Policy used its sandbox in January 2026 to authorize a pilot program with Doctronic, an AI chatbot that reviews and approves prescription refills inside a defined formulary of roughly 190 chronic-maintenance drugs, without a patient seeing a physician first. The state's official agreement with Doctronic defines the eligible medication list, oversight duties, and termination conditions for the pilot.
Today the bot's refills are still reviewed by a human prescriber before the order is sent to the pharmacy. Doctronic says the goal is to move to fully automated refills inside the same list, with a physician only supervising the program rather than each prescription. Co-founder Adam Oskowitz positions Doctronic as handling renewals, not new diagnoses.
In March 2026, eleven of the twelve members of Utah's medical licensing board, including chair Alan Smith, sent the governor a letter urging that the program be paused. Their objection was procedural: the board learned about the pilot from news coverage, after the agreement was already signed, and was not consulted on scope-of-practice questions. The American Medical Association has separately opposed the program on similar scope-of-practice grounds.
Neither the board nor the AMA is claiming the chatbot has hurt anyone. Their argument is that a licensed physician's clinical judgment belongs to the prescription, regardless of who, or what, generates it. Doctronic has cited an 80% diagnostic-match figure from its own internal evaluation, which the company says demonstrates triage accuracy. That number has not been peer-reviewed and measures triage, not prescribing safety, so it is not a benchmark for safe autonomous refills.
The sandbox was meant to be one state. Now it is becoming a template. The Cicero Institute, a libertarian legal-policy group run by Adam Meier, has drafted a model bill that other states can introduce to create their own AI prescribing sandboxes. As of July 2026, no other state has formally adopted it, but the template is the policy product. State legislatures that adopt the template would inherit Utah's scope question without revisiting the underlying clinical question, because the sandbox, by design, is the answer to that question.
An FDA spokesperson told the AP the agency is monitoring the Utah pilot but has not asserted direct jurisdiction over state AI prescribing sandboxes. That leaves malpractice liability, federal preemption, and how state medical practice acts interact with AI agents open. Eric Bressman, a University of Pennsylvania physician, and Daniel Aaron, a University of Utah law professor, both raised those questions when asked about the program.
The next decision belongs in readers' own states. Ask your state medical board whether it has been consulted on any AI prescribing pilot. Ask your pharmacist whether a refill can be issued without a clinician in the loop. The Utah sandbox is running; the question now is who copies it.