Utah has extended its state artificial intelligence initiative into healthcare, introducing pilot programmes that allow software to prescribe medication without real-time human oversight. The Utah Department of Commerce has launched a year-long trial in which consumers can obtain AI-generated prescriptions for topical acne medication. The scheme marks a significant change in how routine prescribing is regulated, because it gives an AI system the authority to assess patients and send prescriptions directly to pharmacies for a limited set of conditions.

The headline programme belongs to Nolla Health, a digital dermatology provider that focuses on mild-to-moderate acne. Patients use the company's Skin App to scan their face and enter their medical history. The software analyses the skin, including its oiliness, and produces a numerical severity score. It then selects a treatment from a list of eight pre-approved medications. The prescription goes straight to a pharmacy, which sidesteps state requirements that ordinarily demand the signature of a human clinician. The exemption applies only within the pilot.

Human involvement diminishes as the trial proceeds. For the first 100 patients, a physician reviews every prescription before it reaches a pharmacy. For the next 400, the AI submits prescriptions directly and physicians examine them retrospectively once a week. After that point the system operates independently, and physicians audit 10 per cent of prescriptions each month. Once the first 500 patients have been treated, nine in every ten prescriptions will therefore be issued and dispensed without a doctor having looked at them.

Nolla Health is one of three AI-based healthcare pilots announced by the state. The Office of AI Policy has placed the other two inside its regulatory sandbox, a framework that lets companies test emerging technologies under temporary exemptions from existing rules. Expect Fitness will provide virtual physical therapy to postpartum mothers, using AI and telehealth tools to allow a small group of clinicians to look after a larger number of patients. August AI will run a prescription renewal system that sorts medications by risk. Lower-risk drugs will skip routine manual approval, while higher-risk treatments stay under stricter oversight.

The state has also signed umbrella agreements with University of Utah Health and Intermountain Health, two major regional healthcare systems. These agreements are intended to speed up the evaluation of future AI deployments within those organisations, which could shorten the path from pilot to routine clinical use.

State officials describe the health sector addition as part of a "pro-human" policy that aims to encourage responsible innovation while protecting public safety. Independent reviewers, among them the Stanford Clinical Excellence Research Center, will assess whether the pilots are clinically useful and whether they protect patients and consumers. Their findings are likely to shape whether the exemptions are extended or narrowed once the trial period ends.

The announcement follows an earlier episode that shows how contested the area remains. Earlier this year Utah introduced a chatbot built on technology from Doctronic, which allowed patients to renew prescriptions without a physician's oversight. The effort was curtailed several months later, after the state's medical licensing board objected to the AI-powered refill management system. The board said it had not been made aware of the application until after it was introduced, and it raised concerns about the lack of advance notice and oversight.

That history gives the new pilots a different footing. Each is structured around a defined number of patients, a published schedule for reducing physician review and a named external evaluator. Whether the licensing board is satisfied with those arrangements has not been made clear in the announcement, and its position will matter given its earlier intervention.

The pilots will provide some of the first real-world data on how far AI systems can be trusted to prescribe without a clinician checking each decision. Acne treatment was chosen as the first test, a condition that is common, visible and treated with topical medication. The findings could guide regulators elsewhere on whether to grant AI systems prescriptive authority for other medical conditions.

‍