Utah regulators announced on Monday a plan that would allow an artificial intelligence product to evaluate patients and write new drug prescriptions with no prior human review — a first-of-its-kind step in the state's expanding health AI regime.

The product comes from Nolla Health, a startup that raised $4.5 million last year from investors including General Catalyst. Nolla will launch an AI-powered service in Utah that assesses patients seeking treatment for mild-to-moderate acne. The system uses a structured intake questionnaire plus analysis of photos of the patient's skin, and then determines whether one of a list of low-risk drugs is appropriate.

During the first stages of the pilot, a licensed clinician will review the AI's decisions and sign off on each prescription. But Nolla may eventually be allowed to prescribe drugs without that human review — the milestone Utah's announcement explicitly contemplates.

What does the sandbox waiver actually change?

The pilot operates under Utah's "AI sandbox" program, which lets the state Office of Artificial Intelligence Policy (OAIP) waive regulations for companies testing AI products under state supervision. The Nolla deal is part of a significant expansion of that program, which regulators and industry watchers have tracked closely as a template for how states can supervise clinical AI without federal frameworks in place.

For R&D and regulatory teams at digital health companies, the structure matters as much as the indication:

• The AI performs the clinical assessment — not a decision-support layer for a physician.
• Prescribing authority moves, in stages, from clinician countersignature to autonomous AI decision.
• The drug list is restricted to low-risk dermatology products for acne, limiting exposure if the model errs.
• The state, through OAIP, acts as the supervising regulator with waiver power.

How credible is the clinical claim?

The source material gives no performance data: no sensitivity or specificity figures for the photo-based skin assessment, no sample size from any validation cohort, and no peer-reviewed results. The severity classification — mild-to-moderate acne — is itself an assessment task where measurement error directly changes drug choice. Nolla's investors include General Catalyst, which has publicly backed the company's ambition to direct treatment for a host of conditions using AI, so the pilot's commercial incentives are visible.

Clinician-in-the-loop review during early stages provides one safeguard, but Utah has not yet published criteria for when or how Nolla would qualify to drop that review. Portfolio managers evaluating autonomous-prescribing AI should treat this as a supervised experiment with undefined graduation thresholds, not a validated clinical pathway.

What comes next?

Utah's expansion signals that state-level sandboxes are becoming the operative venue for autonomous clinical AI in the absence of federal rules, and Nolla's acne pilot will be an early test of whether photo-based AI assessment can carry prescribing authority on its own.