Utah lets Nolla Health's AI write acne prescriptions in pilot

Utah residents can now receive an AI-generated prescription for acne treatment without a live doctor visit, after healthcare startup Nolla Health switched on the service on Monday.

Users scan their faces inside the Nolla Health app, the company's AI system scores acne severity, and the software then writes a prescription on its own, according to a company announcement reported earlier by Bloomberg. Two physicians still sign off on each prescription during the first 100 patients, marking the most cautious phase of a staged rollout that scales back human review as enrollment grows.

The launch lands in a state that has spent the last two years building one of the country's most permissive frameworks for algorithmic medicine. It also gives Nolla Health, a small startup, an early look at how acne — one of the most common dermatological complaints in the United States — responds to a model that treats prescriptions as a software output rather than a clinician's judgment.

How does the Nolla Health prescription actually work?

A patient in Utah opens the Nolla Health app and uses the phone's camera to capture images of their face. The app's AI model reads the photos, classifies acne severity, and generates a treatment plan that includes a prescription. The system is built around what Nolla describes as a fully autonomous prescribing pipeline, with physicians inserted as checkpoints rather than as the originating decision-maker.

The Verge reported the launch on Monday and pointed to Bloomberg's earlier coverage of the same announcement.

How is physician oversight structured across the rollout?

The pilot is split into three review tiers that ease back as the patient count climbs:

• 0 to 100 patients. Two physicians approve every AI-generated prescription before it is issued to the patient.
• 101 to 500 patients. Prescriptions are issued first; physicians review them afterward.
• Beyond 500 patients. Physicians review a sample of at least 10% of prescriptions, according to the company's description of the program.

That stepped-down design turns the first 500 Utah users into a structured safety study. Every prescription in tier one gets a paired human check; tier two shifts to retrospective audit; tier three moves to spot-checking. Nolla has not disclosed which acne medications the system is authorized to prescribe or whether the prescriptions cover topical retinoids, oral antibiotics, or isotretinoin — the latter being tightly controlled under the FDA's iPLEDGE program in the United States.

Why Utah?

Utah has positioned itself as the most AI-friendly state for clinical software. Lawmakers there passed legislation in 2024 and 2025 that explicitly allows AI systems to issue prescriptions under defined physician-supervision rules, a framework that does not yet exist in most other states. The design fits Nolla's pilot neatly: a narrow indication, a defined patient cap, and a regulatory environment that already accepts an algorithm as a prescribing party.

The state-by-state approach is creating a patchwork. California and New York continue to require a licensed clinician to author every prescription, while Texas and Tennessee have passed narrower AI-in-medicine statutes. For Nolla, picking Utah means operating inside the rules rather than fighting them.

What are the stakes for the acne market?

Acne is a $10 billion global treatment market and the second-most-common skin condition in the United States after eczema, according to industry trackers. Most prescriptions route through dermatology, where average wait times exceed 30 days in major metropolitan areas. An app-based pipeline compresses the path from scan to script to a single session.

That compression is exactly what worries clinical groups. The American Academy of Dermatology has warned that acne severity scoring is harder than it looks: borderline lesions can shift a patient between treatment categories, and the wrong topical or oral agent can produce months of side effects. Removing the clinician from the front of the loop trades those risks for speed and access.

Nolla's founder has framed the product as a way to reach patients who never make it to a dermatologist. The company has not yet published peer-reviewed data on its model's accuracy against board-certified dermatologists, which is the comparison that regulators, payers, and clinicians will eventually demand.

What changes at 500 patients?

The transition from tier two to tier three is the regulatory hinge of the rollout. Once Nolla crosses 500 Utah users, the company drops to a 10% sample audit — the level of oversight more commonly associated with radiology AI deployed inside a hospital than with a system that writes prescriptions directly to a pharmacy. From that point forward, 90% of prescriptions issued by the app will not be reviewed by a human before the patient fills them.

If the tier-three audit flags consistent issues, the company has not yet disclosed whether the rollout pauses. If it does not, Nolla gains a dataset and an operating record that other states will struggle to ignore when they draft their own AI-prescribing laws.

Where does this fit in the broader AI-clinic race?

Nolla is not alone. Larger telehealth players have tested AI intake tools that draft notes for a clinician to sign, and a handful of weight-loss startups now use language models to screen eligibility for GLP-1 prescriptions. Nolla's pitch is sharper: skip the clinician as the author of the order entirely, and treat the physician as a quality-control layer that thins over time.

That distinction matters for the FDA, which still treats prescription authority as a clinician-led act under federal law. State frameworks can authorize the practice, but the federal definition of a valid prescription has not been rewritten to cover an algorithm as the prescriber of record. Nolla's Utah pilot runs inside that gap — legal under state law, untested under federal review.

What to watch next

Three milestones will define whether the model spreads past Utah:

• The 500-patient audit. Whether Nolla's 10% sample review surfaces enough errors to slow the rollout or to seed a peer-reviewed paper.
• Federal posture. Whether the FDA opens a formal file on AI-issued prescriptions or stays quiet and lets state law define the practice.
• Adjacent indications. Whether Nolla expands beyond acne into rosacea, eczema, or contraceptive refills — the next three categories where the same scan-plus-prescribe loop is technically feasible.

The first 100 patients in Utah will receive prescriptions that two doctors have approved. By patient 501, nine out of ten of them will not. That ratio is the experiment the rest of the country is about to watch.