Utah has become the first US state where an artificial intelligence system can examine patients and prescribe medication without a doctor directly approving each decision. The program, built around the dermatology startup Nolla Health, went live this week and immediately drew national attention, with the Seattle Times summarizing the milestone bluntly: AI has replaced a prescribing doctor for the first time — to treat acne.
The pilot is deliberately narrow, but its regulatory significance is not. For years, AI in medicine has been positioned as a helper that drafts notes or flags findings for a human clinician to sign off on. Utah's arrangement removes that sign-off for a defined class of patients, making it the clearest test yet of whether regulators, physicians, and patients will accept an algorithm as the prescribing party. For more context on this story, see our ongoing latest AI developments.
How the Nolla Health Pilot Works
The program runs through Nolla Health's app, which costs $4.99 per month and is limited to patients with mild-to-moderate acne. The treatment window is one year.
A patient starts by scanning their face with the app and completing a medical history and personal information form. The AI then analyzes the skin images, rates oiliness, and generates a numerical acne severity score that ranges from clear skin to severe acne, according to TechSpot's report on the launch.
That score is combined with the patient's medical information to produce a prescription recommendation drawn from a list of eight approved treatments. All eight are topical medications rather than pills, a constraint that keeps the most dangerous classes of side effects off the table.
Nolla Health CEO Luis Wenus said the adverse effects associated with the eight approved medications are limited to localized skin irritation or dryness — a materially lower risk profile than systemic drugs, and one of the reasons regulators were apparently willing to let the AI operate with reduced oversight.
A Three-Phase Phase-Out of Human Review
What makes the Utah program notable is not the AI's diagnostic task, which is bounded and visual, but the scheduled retreat of the physician from the loop.
In the first phase, the first 100 prescriptions must still be reviewed by actual doctors before they are submitted to the pharmacy. In the second phase, the next 400 patients have their prescriptions submitted to pharmacies by the AI itself, with a physician reviewing them retrospectively on a weekly basis. In the final phase, a doctor reviews at least 10 percent of all prescriptions issued, checked once a month.
By the end of the pilot, a prescription can travel from a smartphone camera to a pharmacy with no prospective human review at all — only statistical sampling after the fact. Utah's approach effectively treats prescriptions like outputs in a quality-control process rather than individual clinical decisions, a structural shift that medical commentators have long warned would arrive faster than many expected.
Why Acne, and Why Utah
Acne is in many ways the ideal proving ground for autonomous prescribing. It is visually diagnosable, chronic rather than life-threatening, and treated with a small, well-characterized formulary. The downside risk of a wrong choice within the approved list is low and largely reversible, which is precisely what a regulator needs when designing a bounded experiment.
Utah, meanwhile, has spent recent years positioning itself as one of the most AI-forward state governments in the country. The state was previously reported by Bloomberg Law to be weighing its "AI doctor" prescription pilot against oversight concerns, and local outlets including TechBuzz News have described a pro-human AI initiative that recently added a healthcare pillar with named independent evaluators. Trade publication STAT reported that Utah is pressing ahead with additional health AI pilots covering prescriptions and women's health, while MobiHealthNews confirmed Nolla Health's Utah launch this week.
In other words, the acne pilot is not an isolated stunt. It is the most visible piece of a deliberate state-level strategy to move first on autonomous medical AI and build the evaluation infrastructure around it.
The Oversight Debate
Patient-safety advocates have raised familiar concerns. Retrospective review of prescriptions means errors are caught after a patient has already received the medication, not before. Weekly and monthly sampling windows leave room for systematic errors — the kind that affect whole categories of patients — to repeat before a human notices.
There is also the question of precedent. If autonomous prescribing is accepted for acne, the argument for extending it to other visually diagnosable, formulary-bound conditions — conjunctivitis, certain rashes, antifungal treatments — writes itself. Each extension would be individually defensible while collectively moving medicine toward a model where the algorithm is the default prescriber and the doctor is the auditor.
The medical community's unease is compounded by a broader record of consumers relying on general-purpose chatbots for medical advice with harmful results, cases that AI companies themselves have scrambled to constrain. A purpose-built system with a fixed formulary and a defined scoring rubric is a very different artifact from an open-ended chatbot, but the public tends to file both under the same mental category.
What Comes Next
The pilot's one-year window means Utah will have real-world data on autonomous prescribing well before most states have even drafted a position on the question. The numbers that matter are error rates in each phase, patient outcomes across the 500 or so supervised prescriptions, and whether the retrospective review system actually catches the mistakes it is designed to catch.
For now, the program remains what its designers intended: a small, carefully fenced experiment. But the direction of travel is unmistakable, and Utah has just become the first jurisdiction in the United States to find out what happens when the prescription pad is an algorithm.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
