AI Just Wrote Its First Prescription: Utah Lets Nolla Health’s App Start Acne Treatment Without a Doctor Visit (AI Trends, 6 October 2026)

admincybersecurityTagged , , , , , , , , ,

What’s trending in AI on 6 October 2026: For the first time in the United States, a state regulator has authorized an AI system to write a brand-new prescription, not just renew an old one. On 5 October, Nolla Health switched on AI-issued acne prescriptions for adults in Utah under a supervised pilot run through the state’s AI regulatory sandbox. A patient answers structured questions, takes a five-angle face scan and can have a topical treatment prescribed within about 15 minutes, for $4.99 a month. Doctors check every prescription at first, then step back in stages until they review a sample. It is a small product with a big precedent. Below: how Nolla’s system works, the guardrails, what the earlier Utah pilots showed, the objections, and a staged-autonomy scorecard any business can use before letting AI decide on its own.

Key takeaways

  • A first, but a narrow one. Nolla’s AI can start acne treatment for Utah adults, choosing only from a short list of topical creams and gels. Oral drugs such as isotretinoin are excluded.
  • Oversight fades in stages. Two physicians approve every prescription for the first 100 patients; later, doctors review after the fact; eventually they sample at least 10% a month. Each step needs 95% doctor agreement and zero serious adverse events.
  • It is a sandbox, not an approval. Utah’s agreement relaxes specific rules for up to 12 months and says plainly that taking part is not an endorsement.
  • Utah is building a pattern. Doctronic renews chronic medications, Legion Health refills psychiatric drugs, and two more health pilots were signed on 2 October.
  • The fight is about who is accountable. Utah’s medical licensing board, legal scholars and the AMA question whether software should replace clinical judgment, and whether the FDA should review it as a device.
AI just wrote its first prescriptionTitle card. Headline: AI just wrote its first prescription. Subhead: Utah lets Nolla Health’s app start acne treatment without a doctor visit. Three tags: first initial prescription, not a refill; topical acne treatments only; doctors review in stages. Illustration of a phone showing a face-scan outline next to a prescription slip marked AI-initiated. 5-angle scanAnalyze RxAI-initiated AI TRENDS · 6 OCTOBER 2026 AI just wrote its first prescription Utah lets Nolla Health’s app start acne treatment without a doctor visit, under a supervised pilot. First initial prescription, not a refill Topical acne treatments only Doctors review in stages Sources: Nolla Health, Utah Department of Commerce, SiliconANGLE (5 Oct 2026)delana.co
A phone, a face scan and a short list of creams: the first state-authorized AI prescriber in the US starts small.

1. What Utah approved, exactly

Nolla Health, the company behind the Nolla Derm app, signed a regulatory mitigation agreement with Utah on 22 September 2026, according to the state’s list of authorized AI pilots. The agreement is with Utah’s Office of Artificial Intelligence Policy and the Division of Professional Licensing, and it covers a 12-month pilot. The service went live on 5 October. Nolla and Becker’s Hospital Review both describe it as the first time a US state regulator has authorized AI to issue an initial prescription rather than a refill.

The scope is deliberately tight. The pilot is open to Utah residents aged 18 and over. The AI can only choose among physician-approved topical treatments: Unite.AI lists tretinoin, adapalene, clindamycin with benzoyl peroxide, benzoyl peroxide and azelaic acid, plus some compounded versions. SiliconANGLE reports that isotretinoin and all oral medications are off the table. The pilot costs patients $4.99 a month, against the $150 to $300 Nolla says an in-person visit can cost.

Utah’s sandbox does not license the AI or approve the product. The state’s own explainer puts it this way: “An agreement is not permission to do whatever the company likes.” It is a narrow, written, time-limited change to how certain rules apply, with monthly reporting and audits. Pilots can run for up to 12 months, with up to two 12-month extensions. Prescriptions still go out under an overseeing physician’s credentials, and Unite.AI reports they are labeled as AI-initiated.

2. How an AI prescription works, step by step

According to Nolla’s own description, the patient downloads the app, verifies their identity, gives informed consent, answers intake questions and takes a face scan from five angles. The system scores the severity of the acne and builds a treatment plan. The whole process takes 10 to 15 minutes. The prescription can be picked up at a local pharmacy or delivered. After that, daily skin scans track progress, and the plan is reviewed after every 28 scans, roughly every one to two months.

Two design choices matter more than the AI model itself. First, there is no free-text chat: patients answer structured questions, which keeps the system inside its protocol and leaves little room for the kind of prompt manipulation that has tripped up chatbots elsewhere. Second, there are hard stops. Unite.AI reports that a severity score above 3.5 on a five-point scale sends the case to a physician, a score of 4.0 directs the patient to in-person dermatology care, and pregnancy, immunosuppression, kidney or liver disease and medication allergies all block access. Messages from patients are answered by licensed physicians, not by the AI.

From face scan to prescription in about 15 minutesFlow diagram of Nolla Derm’s pilot. Step 1 verify identity. Step 2 informed consent. Step 3 structured intake questions with no free text. Step 4 five-angle face scan. Step 5 AI severity score and plan. Step 6 topical prescription, pharmacy pickup or delivery. A branch in orange shows hard stops: severity above 3.5 goes to a physician; 4.0 means in-person dermatology; pregnancy, immunosuppression, kidney or liver disease and allergies block access. From face scan to prescription in about 15 minutes How Nolla Derm’s Utah pilot handles a new patient 1Verify identity 2Informed consent 3Structured intake 45-angle face scan 5AI severity score 6Topical Rx no free-text chat Hard stops Severity above 3.5 → physician reviews the caseSeverity 4.0 → referred to in-person dermatologyPregnancy, immunosuppression, kidney/liver disease, allergies → blocked After the first prescription Daily skin scans track progressPlan reviewed every 28 scansPatient messages answered by physicians Sources: Nolla Health; Unite.AI; Becker’s Hospital Reviewdelana.co
The guardrails are as much about what the AI cannot hear or decide as what it can.

3. The guardrail ladder: how doctors step back

The most transferable idea in this pilot is not the face scan. It is the way human oversight is withdrawn gradually and only when the numbers justify it. Nolla’s published plan has three stages:

  1. Stage 1 (at least four weeks, first 100 patients): two physicians independently review and approve every AI prescription before it reaches the patient.
  2. Stage 2 (at least eight weeks, up to 500 patients): the AI issues prescriptions directly, and physicians review all cases after the fact.
  3. Stage 3: physicians review at least 10% of prescriptions each month, plus every escalation and every reported side effect.

Moving up a stage requires 95% physician agreement with the AI, zero serious adverse events and the state’s sign-off. Reports differ slightly on the review rhythm in later stages, so treat Nolla’s own description as the reference. Unite.AI adds that Nolla must report case counts, prescriptions by drug, clinician agreement rates and patient demographics every month, that quarterly reports will be made public, and that third-party auditors will stress-test the guardrails.

On performance, Nolla says licensed clinicians agreed with its treatment choice in about 96% of its most recent 1,000 real-world cases, with disagreements limited to adjustments such as strength or dosage within the same drug class. It also reports that its severity model scored an R-squared of 0.947 on 10,348 test cases, after training on more than 100,000 labeled cases. These are company figures, not independent results; the public quarterly reports will be the real test.

The guardrail ladderThree-step ladder showing how physician oversight is reduced in Nolla’s pilot. Stage 1, at least four weeks and the first 100 patients: two physicians approve 100 percent of prescriptions before release. Stage 2, at least eight weeks and up to 500 patients: AI prescribes and physicians review 100 percent afterwards. Stage 3: physicians sample at least 10 percent monthly plus all escalations and side effects. Gate between stages: 95 percent physician agreement, zero serious adverse events and state approval. The guardrail ladder How much of the AI’s work a physician checks at each stage of Nolla’s pilot 100%approved beforerelease, by two doctors 100%reviewed afterthe AI prescribes ≥10%monthly sample + escalations Stage 1 · first 100 patientsat least 4 weeksStage 2 · up to 500 patientsat least 8 weeksStage 3 · ongoingwith third-party audits GATEGATE 95% agreement0 serious eventsstate approval95% agreement0 serious eventsstate approval Source: Nolla Health pilot description (5 Oct 2026); Unite.AIdelana.co
Autonomy is earned with data at each gate, not granted on launch day.

4. Why acne, and why Utah

Acne is close to an ideal first case for autonomous prescribing. It is common, visible on camera, usually treated with well-understood topical drugs and rarely an emergency. Nolla’s numbers on the access gap are stark: an average wait of 61 days for a dermatologist in Utah against a national median of 32.3 days, 11 Utah counties with no dermatologist at all, and only around 10% of the roughly 100,000 Utahns with active acne seeing a specialist. Nolla says its app, which works under clinician guidance in 44 states, has more than 175,000 downloads; AI-only prescribing applies in Utah alone.

Nolla’s chief medical adviser, Zaid Fadul, told SiliconANGLE that acne is a starting point, with flu and uncomplicated urinary tract infections named as possible next conditions. He was equally clear about where the line sits: stimulants, opioids and benzodiazepines are poor candidates for autonomous prescribing because patients may try to game the system. “The key is having very clear guardrails,” he said.

The access gap AI is trying to closeBar chart comparing average dermatology wait times: 61 days in Utah versus a 32.3-day national median. Side panels: 11 Utah counties have no dermatologist; only about 10 percent of Utahns with active acne see a dermatologist; an in-person visit costs 150 to 300 dollars versus 4.99 dollars a month for the AI pilot. The access gap AI is trying to close Average wait for a dermatology appointment, in days 61Utah average 32.3US median 11Utah counties withno dermatologist ~10%of Utahns with active acnesee a dermatologist $4.99a month in the pilot, vs$150–$300 for a visit Figures cited by Nolla Health, via Unite.AI and Becker’s Hospital Reviewdelana.co
The business case for AI prescribing is an access case: long waits and empty counties.

5. Utah’s growing AI-in-medicine sandbox

Nolla is not Utah’s first health pilot, and it will not be the last. The state’s list of authorized pilots shows a steady expansion from refills toward new decisions:

PilotWhat the AI doesAgreement signedKey limits reported
DoctronicRenews prescriptions for chronic conditions24 Oct 2025About 190 non-controlled drugs; no controlled substances, ADHD drugs or injectables; doctors review the first 250 decisions per drug class
Legion HealthRefills psychiatric medications for stable patients19 Mar 202612-month pilot
Nolla HealthIssues initial topical acne prescriptions from a photo assessment22 Sep 2026Adults 18+, topical drugs only, staged physician review
August AI (August Labs)Authorizes routine medication refills2 Oct 2026Details not yet widely reported
Expect FitnessAI-assisted pelvic floor physical therapy plans2 Oct 2026Details not yet widely reported
University of Utah Health and Intermountain HealthMaster agreements for multiple future pilotsSep 2026Individual pilots to follow
Compiled from the Utah Department of Commerce list of authorized pilots, Medical Economics and Becker’s Hospital Review.

The earlier pilots give a first glimpse of how autonomous medical AI behaves in practice. Becker’s reports that in Doctronic’s early months the AI recommended renewal in 72% of cases and physicians agreed with 91% of those recommendations. The other 28% were escalated to humans, and doctors judged nearly a third of those escalations overly cautious. The system leaned toward asking for help: the right direction to fail in, but a reminder that “autonomous” systems still create human workload. Becker’s adds that Doctronic has held talks about expanding to Arizona, Missouri, Texas and Wyoming.

From refills to first prescriptions: Utah’s timelineTimeline. October 2025: Doctronic prescription renewal pilot agreement. March 2026: Legion Health psychiatric refills approved. April 2026: Utah Medical Licensing Board asks for the refill pilot to be suspended. June 2026: AMA adopts policies opposing AI replacing physician judgment. 22 September 2026: Nolla Health agreement signed. 2 October 2026: August AI and Expect Fitness pilots signed. 5 October 2026: Nolla’s AI-issued acne prescriptions go live. Teal marks pilot approvals, orange marks pushback. From refills to first prescriptions Utah’s AI health sandbox, and the pushback, 2025 to 2026 Oct 2025Doctronicrenewals Mar 2026Legion Healthpsych refills Apr 2026Licensing boardasks to suspend Jun 2026AMA policy onphysician judgment 22 Sep 2026Nolla agreementsigned 2 Oct 2026August AI,Expect Fitness 5 Oct 2026First AI-issuedprescriptions live Teal: pilot approvals and launches Orange: pushback from the medical profession Sources: Utah Department of Commerce; Becker’s Hospital Review; Bloomberg Law; Nolla Healthdelana.co
Each pilot pushes the boundary a little further, from renewing old decisions to making new ones.

6. The objections: safety, the FDA and who is accountable

The pushback is real and has been building since the refill pilots. Bloomberg Law reports that members of Utah’s Medical Licensing Board wrote in April 2026 that automating prescriptions without professional review could compromise patient safety, complained they had not been consulted, and asked for the refill pilot to be suspended. Legal scholars have raised a federal question: Harvard’s Aaron Kesselheim argued Doctronic’s technology should be regulated as a medical device, and University of Illinois law professor Sara Gerke asked why the FDA has not reviewed it, given the potential for harm. Becker’s notes that on 10 June 2026 the American Medical Association adopted policies opposing AI replacing physician judgment.

Three practical issues sit underneath the debate, and each applies far beyond medicine:

  • Accountability. Prescriptions go out under a supervising physician’s name. If an AI-initiated prescription harms someone, the doctor, the company and the state’s agreement all sit in the chain. We explored the same question for software agents in “The AI Did It” Is No Longer a Defense.
  • Privacy and security. Face scans, identity documents and medical histories are among the most sensitive data a company can hold. Nolla says it is HIPAA and GDPR compliant, but a daily-scan model creates a large, ongoing biometric data stream. Health data breaches are already a live risk; see our report on the AI agent behind the Australian Medicare breach.
  • Drift and gaming. A system that performs well on 1,000 cases can behave differently on the next 10,000, and patients learn which answers work. Ongoing sampling matters more than launch-day accuracy.

The supporters’ case is just as concrete: a 61-day wait for a common, treatable condition is also a harm, and Utah’s model replaces vague promises with gates and audits that can be checked.

7. What it means for businesses outside healthcare

Most Delana readers will never prescribe anything. But almost every business is now deciding where AI can act without a human: approving refunds, setting credit limits, screening job applicants, closing support tickets, pushing code. Utah’s pilot is a public, documented template for exactly that decision, and it is more disciplined than what most companies do. Most teams test an AI tool, like the results and switch it on for everyone. Nolla’s pilot limits scope, starts with full review, sets numeric gates and commits to audits.

Regulators are moving in the same direction. Sandboxes like Utah’s are spreading, and Doctronic’s expansion talks suggest other states may follow. If you sell AI-driven services, expect customers and regulators to ask for the same evidence: a defined scope, escalation rules, agreement rates and incident counts. If you buy them, ask for it first. Our vendor questions guide is a good starting list, and our older piece on how much autonomy to grant AI covers the basics. The insurance angle matters too: many policies now carve out AI decisions, as we explained in Cyber Insurance and AI Exclusions in 2026.

8. The staged-autonomy scorecard: before AI decides on its own

Borrow Utah’s ladder for any AI that makes decisions in your business. Score each line 0 (not in place), 1 (partly) or 2 (fully, with evidence). Then use the total to decide how much human review the system still needs.

#GuardrailWhat “2 points” looks likeNolla’s equivalent
1Narrow scopeA written list of the decisions the AI may make, and everything else is excludedTopical acne drugs only; no oral medication
2Structured inputsThe AI acts on forms and fields, not open-ended text that could carry hidden instructionsNo free-text interface
3Hard stopsClear conditions that always send the case to a personSeverity thresholds, pregnancy, organ disease, allergies
4Full review to startA person approves every decision for a defined first batchTwo doctors approve the first 100 patients
5Numeric gatesAgreement rate and incident thresholds written down before launch95% agreement, zero serious adverse events
6Ongoing samplingA fixed share of decisions reviewed forever, not just at launchAt least 10% a month plus all escalations
7Labeled decisions and a named ownerEvery AI decision is marked as such and a named person is accountablePrescriptions labeled AI-initiated, under a physician’s credentials
8Outside reporting and auditRegular reports and an independent check of the guardrailsMonthly state reports, public quarterly reports, third-party auditors
Delana’s scorecard, modeled on the oversight structure in Utah’s Nolla Health pilot.

How to read your score: 0 to 7 means keep a human approving every decision; the AI is an assistant, not a decision-maker. 8 to 12 means the AI can act, but a person should review every case after the fact until the gaps are closed. 13 to 16 means you can consider moving to sampling, as long as the sample never drops to zero. If you cannot score line 1 at all, start with an inventory: our shadow AI guide helps uncover the AI decisions your teams already make without approval, and AI Agent Security in 2026 covers who should own them.

9. What to watch next

Four things: Nolla’s first public quarterly report, which will show whether the 96% agreement figure holds once the AI prescribes on its own; whether the pilot advances from Stage 1 to Stage 2 on schedule; any FDA statement on whether autonomous prescribing software is a medical device; and whether Doctronic’s expansion talks, or Nolla’s next targets of flu and urinary tract infections, reach other regulators.

Frequently asked questions

Can AI legally prescribe medication in the US now?

Only in narrow, supervised pilots. Utah’s AI sandbox has let Nolla Health’s AI issue initial prescriptions for topical acne treatments to adults in Utah since 5 October 2026, under a 12-month regulatory mitigation agreement. Prescriptions go out under an overseeing physician’s credentials. This is general information, not legal or medical advice.

What can Nolla Health’s AI prescribe?

Only physician-approved topical acne treatments, such as tretinoin, adapalene, benzoyl peroxide, clindamycin with benzoyl peroxide and azelaic acid. Isotretinoin and all oral medications are excluded, and severe acne, pregnancy and certain health conditions send patients to a physician instead.

Do doctors still check the AI’s prescriptions?

Yes, in stages. Two physicians approve every prescription for the first 100 patients. Then the AI prescribes directly and doctors review every case afterwards. Finally, doctors review at least 10% each month plus all escalations and side effects. Each step requires 95% physician agreement, zero serious adverse events and state approval.

What is Utah’s AI regulatory sandbox?

It is a program run by Utah’s Office of Artificial Intelligence Policy that gives companies narrow, time-limited relief from specific rules so they can test AI products under supervision, with monthly reporting and audits. Pilots last up to 12 months with up to two extensions. Utah says taking part is not an endorsement of the product.

How is this different from Doctronic’s AI refills?

Doctronic’s pilot, agreed in October 2025, renews existing prescriptions for about 190 non-controlled chronic-condition drugs. Nolla’s AI starts a new treatment for a patient who has no prior prescription, which is why it is described as the first authorized AI-issued initial prescription in the US.

What are the main concerns about AI prescribing?

Utah’s Medical Licensing Board has warned that automated prescribing could compromise patient safety, legal scholars argue such tools should be reviewed by the FDA as medical devices, and the AMA has adopted policies opposing AI replacing physician judgment. Other concerns include accountability for errors, privacy of face scans and health data, and patients gaming the system.

Sources