Nolla Health will assess Utah patients for mild to moderate acne and write a low risk prescription, with a clinician signing off in the first phase and a path to unsupervised prescribing under the state AI sandbox.
Utah's Office of AI Policy cleared Nolla Health to launch an AI that assesses patients for mild-to-moderate acne, picks from a list of low-risk drugs, and writes the prescription. A licensed clinician will sign off in the first phase. The same state order eventually lets the company drop that step, leaving an AI to prescribe inside a defined clinical lane.
The change sits inside a structure, not a free-for-all. The framework is Utah's AI sandbox, run by the Office of Artificial Intelligence Policy (OAIP), which can waive specific state rules for companies testing AI products under supervision. Each pilot gets its own waiver, its own list of allowed products, and a third-party evaluator to audit outcomes. The state requires signed receipts for designated sandbox systems, an attestation that the AI behaved as expected on a given case.
Nolla's track is narrow by design. The AI takes a structured intake and patient photos, then decides whether one of a list of low-risk drugs is appropriate for mild-to-moderate acne. In phase one, every prescription still goes to a licensed clinician for review. Only later, under the sandbox waiver, would that human step go away. The company says the model is a floor, not a ceiling: Nolla raised $4.5 million from investors including General Catalyst and has signaled plans to extend the same approach to other conditions.
The audit floor is what makes the experiment testable. Vega Health has been selected as a third-party evaluator for the sandbox, the independent check that turns a pilot from announcement into evidence. OAIP's signed-receipt rule means each prescription is logged and attributable. If outcomes drift, the receipts are the rollback mechanism. The clean test case: does the AI hold its safety and accuracy once the clinician sign-off is removed, or does performance degrade in ways the receipts catch?
The sandbox is not just an acne story. On the same day, August AI was approved to pilot AI-assisted prescription renewals for patients with chronic conditions in Utah, a different clinical lane with refill decisions and a different risk profile. STAT+ also reports the state is expanding into women's health pilots. Each pilot is a separate waiver inside the same oversight system, which lets Utah test more capability under more conditions without picking one model for the whole.
For a patient in Utah, the practical effect is access. A mild-to-moderate acne case today often means a dermatologist wait, a copay, and a follow-up visit for what is usually a standard topical. If the Nolla pilot holds, that same patient could send photos and an intake form, get a clinical decision in minutes, and pick up a prescription the same day. The clinician sign-off, the low-risk drug list, and the audit receipts are not friction for friction's sake; they are what makes the wider experiment legible. The next test is whether the receipts stay sharp when the human step goes away.