Utah has become the first U.S. state to authorize artificial intelligence systems to examine patient information and issue prescriptions under a regulatory sandbox program, with pilots that progressively reduce direct human oversight. The Office of Artificial Intelligence Policy (OAIP) within the Utah Department of Commerce approved the arrangements under the state's Artificial Intelligence Policy Act, allowing selected companies to operate with temporary regulatory relief while meeting defined safety and reporting requirements.

The development marks a structured test of autonomous clinical decision-making for routine care. Early programs focused on prescription renewals for chronic conditions. A more recent expansion added authority for initial prescriptions in limited dermatology cases. Officials describe the pilots as a way to improve access, especially in areas with clinician shortages, while collecting performance data before broader deployment.

Medicines
THLT LCX via unplash.com, modified by ChatGPT

Technical Architecture and How the Systems Operate

The primary early pilot partners with New York-based Doctronic. Its platform processes renewal requests for approximately 190 non-controlled medications used for chronic conditions such as hypertension, diabetes, thyroid disorders, and certain maintenance psychiatric drugs. Patients confirm they are physically located in Utah, upload identification, complete a clinical questionnaire, and provide a selfie for identity verification. The system cross-checks pharmacy databases, including Surescripts records, to confirm prior prescriptions and screen for interactions or gaps in therapy.

If the AI determines a renewal is appropriate, it can transmit the order directly to a pharmacy for 30-, 60-, or 90-day supplies. Cases that fall outside predefined parameters or raise uncertainty are escalated to a licensed Utah physician. Controlled substances, ADHD medications, injectables, and certain higher-risk drugs are excluded.

In October 2026 the OAIP expanded the sandbox to include Nolla Health's Nolla Derm application. Adults in Utah with mild-to-moderate acne submit multi-angle facial photographs and medical history. Computer vision algorithms score severity and select from a restricted formulary of topical medications, including some compounded formulations. The system can authorize both initial prescriptions and renewals. August AI received parallel approval for chronic-condition refill processing under similar constraints.

Both platforms operate under phased oversight agreements. For Doctronic, physicians review the first 250 decisions in each medication class before the system may handle subsequent cases with reduced prospective review. Nolla Health begins with physician approval of every prescription for the first 100 patients, shifts to daily retrospective review through the first 500 patients, and later moves to weekly sampling of a subset of cases. Pharmacists retain the ability to flag or refuse any order.

Empirical Results and Operational Scope

Utah Department of Commerce data covering the first months of the Doctronic pilot showed the AI recommended renewal in 72 percent of completed requests and escalated the remainder. Reviewing physicians agreed with 91 percent of the AI's renewal recommendations (97 percent when a second physician opinion was considered). The company previously reported that its treatment recommendations matched human clinicians in 99.2 percent of 500 urgent-care comparison cases. No serious adverse events were reported in the early public summaries.

Service fees have been set low: Doctronic charged approximately $4 per renewal in initial reporting, while Nolla Derm launched at $4.99 per month. Participation requires patients to be Utah residents aged 18 or older for the acne program. The pilots remain limited to the approved formulary and geographic boundaries; expansion beyond Utah requires additional state agreements.

Dr. Adam Oskowitz, co-founder of Doctronic and associate professor of surgery at the University of California San Francisco, stated: "The AI is actually better than doctors at doing this. When you go see a doctor, it’s not going to do all the checks that the AI is doing."

Zaid Fadul, M.D., chief medical adviser for Nolla Health, said of the oversight structure: "The real test for AI in healthcare isn't how smart it is. It's whether someone other than the company can shut it off."

Industry Context and Remaining Challenges

The Utah Medical Licensing Board issued a letter in April 2026 calling for immediate suspension of the Doctronic pilot, arguing that prescription renewal requires ongoing clinical reassessment for side effects, interactions, and continued appropriateness. Eleven of 14 board members signed the recommendation. State regulators continued the program while adjusting the medication list and increasing third-party evaluation requirements in later pilots.

Liability allocation between the AI developer, supervising clinicians, and the state remains unresolved under current sandbox authority. Federal questions about whether the systems constitute regulated medical devices or practice of medicine have also been raised in legal analyses. Over the next 12 to 24 months, the OAIP plans to gather additional safety metrics, refine governance guidance, and consider statutory frameworks for liability. Success criteria include sustained agreement rates with human reviewers, absence of adverse events, and demonstrated improvements in medication continuity.

Utah officials have positioned the sandbox as a controlled environment for testing rather than permanent deregulation. Additional pilots covering physical therapy recommendations and other low-risk clinical tasks have been authorized under the same framework. Outcomes from the current prescription programs will inform whether similar autonomy can be extended to other routine medical decisions.

Sources: