In a modest office building near Salt Lake City, regulators have greenlit something unprecedented. An artificial intelligence system can now examine a patient’s face via smartphone camera, score the severity of their acne and issue a prescription. No doctor visit required. At least not at first.

The company behind it, Nolla Health, launched its Utah-only service this week. For $4.99 a month, adults 18 and older can download the Nolla Derm app, verify their identity, answer a medical questionnaire and submit photos from five angles. The AI analyzes skin condition, oiliness and other factors. It generates an acne severity score. Then it selects from eight approved treatments and writes the script if appropriate. CNET reported the details shortly after launch on Oct. 7.

But this isn’t the first time Utah has handed clinical decisions to code. Months earlier the state partnered with Doctronic, a New York telehealth startup. Its AI chatbot began renewing prescriptions for nearly 190 common medications used to treat chronic conditions. Blood pressure drugs. Antidepressants. Birth control. The list excludes controlled substances, painkillers, ADHD medications and injectables. Patients pay about $4 per renewal. They confirm residency, upload identification and answer questions pulled from a national pharmacy database. The system decides. Or escalates to a human.

Doctronic’s co-founder Dr. Adam Oskowitz, an associate professor of surgery at the University of California, San Francisco, didn’t mince words. “The AI is actually better than doctors at doing this,” he told Politico in January. The company had tested its model against physicians on 500 urgent care cases. Treatment plans matched 99.2% of the time.

Early data from the Doctronic pilot backed some of that confidence. In reviewed cases the AI recommended renewal in 72% of encounters. Physicians agreed 91% of the time, according to reports in Fierce Healthcare. Yet the rollout sparked immediate backlash. The Utah Medical Licensing Board demanded the program be “immediately suspended.” Refills require clinical judgment, board members argued. Side effects can emerge. Drug interactions appear. Patients might stay on outdated therapy for years. MedPage Today covered the clash in April.

Utah didn’t pause. Instead it doubled down. On Oct. 7 state officials announced an expansion of the regulatory sandbox that made these experiments possible. New pilots include August AI for broader chronic condition refills and Expect Fitness for AI-guided pelvic floor physical therapy. Nolla Health’s acne tool joined the group. Officials also created a clinical AI advisory group and lined up independent evaluators such as the Coalition for Health AI and Stanford’s Clinical Excellence Research Center. Fierce Healthcare detailed the updates the same day.

The sandbox works like this. Utah’s Office of Artificial Intelligence Policy, led by Zach Boyd, signs agreements that temporarily waive certain medical licensing rules. Companies must follow strict contracts. Disclose that patients are interacting with AI. Maintain escalation protocols. Submit to audits. The goal is clear. Test bold ideas in a controlled setting. Gather evidence. Shape national policy. Boyd has said Utah wants to become a test bed for responsible AI governance in health care.

Critics remain unconvinced. Physicians worry the technology skips the nuance that comes with years of training. What happens when a patient on antidepressants develops new symptoms the chatbot misses? Or when facial scan quality varies by lighting and skin tone? Liability questions linger. So do concerns about the Food and Drug Administration. Some legal experts argue these AI systems qualify as medical devices that should face federal review. STAT explored that tension in a newsletter published Oct. 7.

Proponents point to access. Rural Utah counties face doctor shortages. Medication non-adherence costs the health system billions and contributes to preventable deaths. A $5 monthly subscription or $4 renewal fee beats missed work for an in-person appointment. Matt Pavelle, Doctronic’s co-CEO, has emphasized easier access for patients who otherwise wait weeks. Nolla Health executives echo the theme. Their system, they say, handles routine low-acuity cases so dermatologists can focus on complex ones.

The oversight ramps down in phases. For Nolla Health’s first 100 prescriptions, two board-certified physicians must review and approve every AI decision before it reaches a pharmacy. Success metrics include at least 95% agreement between AI and reviewers plus zero serious adverse events. Later stages reduce reviews to weekly audits, then 10% sampling. Doctronic followed a similar ladder. Initial batches reviewed in full. Then retrospective checks. Escalation triggers remain active. New symptoms. Conflicting records. Patient requests for human input.

And the results so far? Mixed but instructive. Doctronic’s internal numbers look strong. Physician agreement hovers in the low 90s. Yet the medical board letter highlighted real risks. One physician told CNN in July that the board felt sidelined. “We were essentially told: ‘Yes this is going on. And no, you don’t have a say in it.'” The program continued anyway.

This tension captures a larger shift. Health systems strain under demand. Training pipelines for new doctors lag. AI offers speed and consistency on narrow tasks. But medicine has never been only about narrow tasks. Context matters. Trust matters. A facial scan cannot capture the full patient story the way conversation can.

Still, Utah pushes forward. Its sandbox now includes major players. Intermountain Health and University of Utah Health signed master agreements to test additional AI tools while maintaining data privacy and governance standards. The state requires transparency. Patients must know they face an algorithm. That disclosure itself raises questions. Does knowing change behavior? Will patients trust the output less or more?

Other states watch closely. Arizona, Texas and Wyoming have similar regulatory relief programs. If Utah’s experiments produce clean safety data, pressure will grow to replicate them. If problems surface, lawsuits or federal intervention could follow. The FDA has yet to issue clear guidance on autonomous prescribing AI. That silence won’t last.

For now the pilots remain limited. Acne creams. Statins. SSRIs. Nothing that demands intensive monitoring. The first true test will come as oversight loosens and patient volume scales. Will error rates stay low? Will pharmacists feel comfortable dispensing scripts signed by software? Will patients actually improve adherence?

Utah has placed a bet. The technology is ready enough. The safeguards sufficient. The potential benefits outweigh the risks. Physicians disagree. Regulators in other states hesitate. Yet the experiments continue. Data will accumulate. Debates will sharpen. And somewhere in Salt Lake City a server will keep scanning faces and renewing scripts. Quietly. Efficiently. One prescription at a time.

Whether this marks the start of a safer, more accessible system or a cautionary tale depends on what the audits eventually reveal. For the moment, the code holds the pen. Doctors stand by, reviewing. The question is how long that balance holds.