Table of Contents
AI prescriptions arrived in America on October 5, 2026: a Utah regulatory waiver let Nolla Health’s dermatology app write initial topical prescriptions with no physician real-time signature — and the company’s chief executive told Bloomberg the drafts “carry a clinician’s name without a human professional writing them.” That last detail set off the “no doctors” headlines. The signed agreement on the state’s website tells the real rules behind the headline, and they are more precise: two independent Utah-licensed physicians approve every prescription for the first 100 patients, a licensed reviewer checks 100% of the next 400 weekly, and monthly sampling plus 24-hour adverse-event reporting runs for as long as the waiver lasts. This guide unpacks the actual document — what got approved, the three-stage check ladder, the numeric gates a company must clear before supervision loosens, and who revolted on the way to the signature.
Everything below is sourced from the primary documents (the waiver agreement and the government pilots page), the named outlet reports on it, and the companies’ own announcements — with vendor claims labeled as vendor claims throughout.
Key Takeaway
- 🩺 AI prescriptions, first in the nation: Utah’s Office of AI Policy authorized an AI to write initial prescriptions — topical acne treatments for adults — effective October 5, 2026, inside the state’s AI Learning Laboratory regulatory sandbox.
- 🔁 “No doctors” is the wrong headline: the agreement keeps physicians at every stage — two pre-approve all prescriptions through 100 patients, the next 400 get full weekly retrospective review, then monthly sampling of at least 10%.
- 📊 Advancement is gated, not automatic: at least 95% AI-physician agreement, zero missed hard stops and zero serious adverse events per stage, with the state’s written approval required each time supervision loosens.
- ⚖️ The conflict is the news: 11 of 14 physicians on Utah’s Medical Licensing Board formally sought to stop the earlier AI refill pilot in April — the state refused — and the new pilot only advanced after meeting the board, which added its own appointed physician overseer.
- 🇵🇭 Why it matters beyond one state: the sandbox-plus-gates structure is now the world’s first live template for regulated AI prescribing — the same governance question health systems (including Philippine telehealth under data-privacy rules) will face as AI clinical tools arrive.
What actually got approved: the sandbox rule that lets AI prescriptions happen
Utah did not pass a new medical-practice statute for this. The authority runs through a 2024 law (SB 149) that created an Office of Artificial Intelligence Policy inside the Department of Commerce and, with it, an AI Learning Laboratory — a regulatory sandbox in which a company signs a Regulatory Mitigation Agreement and receives limited, time-boxed relief from specific licensing rules, so it can test a technology under written safeguards. The agreements run twelve months, can be renewed at most twice at the office’s discretion, and can be terminated by the state at any time. The Nolla Health agreement was signed on September 22, 2026, co-signed by the Division of Professional Licensing — the same agency that enforces the rule requiring a licensed physician prescriber.
| Term | What it means |
|---|---|
| Scope: what the AI prescriptions cover | Initial prescriptions AND refills for topical acne treatments (creams and gels from a short physician-set list) — no oral medication, no isotretinoin, no severe acne |
| Who is eligible | Utah adults 18+ with mild-to-moderate acne; identity and age verified by government photo ID; prescriptions ship only inside Utah |
| Hard-stop exclusions | Pregnancy or planning pregnancy, breastfeeding, weakened immune system, or a past reaction to any listed medication stops the process |
| What the pharmacist sees | The prescription is flagged as AI-issued, with a named physician to contact for questions |
| Reporting duties | Adverse events reported to the state office within 24 hours; monthly reports include the rate of AI-physician agreement |
| Insurance | Enrollment and dispensing may not begin until coverage is bound and confirmed to the state |
Safety is the stated argument for the list being short: the company’s chief medical advisor told reporters the AI “can’t improvise — it can only choose from a short list of physician-approved topical treatments.”
The three-stage ladder of AI prescriptions supervision
| Stage / term | Patient volume | Who reviews, and when | Gates to clear |
|---|---|---|---|
| Baseline (12 weeks) | enrollment building | Licensed physicians review, approve and submit — the AI only recommends; concordance reported at 95%+ | Concordance at or above 95% over the preceding 12 weeks |
| Stage 1 (4 weeks) | first 100 patients | TWO independent Utah-licensed physicians approve EVERY prescription BEFORE it is sent | 100% pre-review compliance; zero missed hard stops; zero serious adverse events |
| Stage 2 (8 weeks) | 500 cumulative | The AI submits directly; ALL cases reviewed retrospectively at least weekly (concordance measured on 100%) | Same numeric gates; the state’s prior written approval to advance |
| Stage 3 (750 cumulative) | 750 cumulative | Monthly sampling of at least 10% of enrollment; every adverse event or escalation reviewed in full | Ongoing concordance; monthly reporting kept current |
The line the agreement draws on AI prescriptions — “the company shall not commence any succeeding phase without the prior written approval of the Office” — is the part that separates this from automation without brakes. This is the same governance principle as the sandbox logic inside the sovereign AI race: states run the rulebook, and the company earns every loosening it gets.
The claim check: the real rules versus the “no doctors” stories
What the claim gets RIGHT: the first AI prescriptions without a physician signing in real time — a US-state first; after Stage 1 ends, no physician must read a prescription before the patient’s request reaches the pharmacy (first weekly retrospective, then sampling), and the prescriptions carry a clinician’s name that no human wrote. The LA Times called it “the first time a state has allowed such technology to replace the role of a doctor” — accurate, with the acne-only scope in the same breath.
What the claim gets WRONG: AI prescriptions without any human review. The signed agreement keeps licensed physicians at every stage — universal pre-approval (Stage 1), full weekly retrospective review (Stage 2), and structured sampling with a minimum percentage (Stage 3) — plus hard-stop exclusions, pharmacist disclosure, and state approval for every phase change. The honest summary: the doctor’s real-time signature is what got waived; the doctor’s oversight did not. Utah state officials themselves call it a “narrow exception.”
The doctors’ revolt — and the compromise that followed
The conflict history is not an aside; it explains the shape of the final agreement. In April 2026, 11 of the 14 physicians on the Utah Medical Licensing Board signed a letter calling for the immediate suspension of an earlier pilot — the Doctronic AI-renewal program — citing patient-safety concerns, a lack of clinical oversight and that the board had not been consulted before launch. The state declined the suspension. The earlier refill pilots with Doctronic and Legion Health (routine 30-, 60- and 90-day renewals of existing prescriptions) drew the sharpest institutional resistance because they touched medications beyond a single narrow condition.
Nolla’s route to AI prescriptions differed: the company met the board before authorization and the final program was amended to include a third-party physician appointed by the board to oversee prescription reviews. The contrast — one company igniting the board, the other consulting it — is now a case study in how AI healthcare startups should enter a regulator’s sandbox. Two further pilots signed October 2, 2026 follow the same structure: August AI (routine refills of maintenance medications — provider checks on the first 250 refills, then the next 1,000, then at least 5% sampling, with no more than three AI refills in a row before a human visit) and Expect Fitness (AI-supported pelvic-floor therapy programs).
Why Utah built it this way: the access argument
The company’s pitch is dermatology scarcity, and the numbers are the pitch: Nolla cites about 100,000 Utah residents with active acne at any time, with only about 10% ever seeing a dermatologist; an average dermatology wait of 61 days — nearly twice the national median the company cites — and 11 counties with no dermatologist at all. Acne is the natural first domino for AI prescriptions: it is image-diagnosable (a photo plus an intake questionnaire), self-limited in scope, overwhelmingly treated topically, and already handled largely without primary-care physicians in retail pharmacies. The app behind it (Nolla Derm) claims 175,000+ downloads across 44 states, and the company raised $4.5 million led by General Catalyst; the AI-prescribing pilot itself is Utah-exclusive at the promotional rate of $4.99 a month. All vendor figures carry their labels — the state’s signature is on the oversight, not the market claims.
The federal echo is real but pending: Representative David Schweikert’s Healthy Technology Act of 2025 proposes letting AI systems prescribe FDA-approved medications under defined guardrails — a bill, not law, and Utah’s pilot is now the live experiment legislators will point to in both directions.
WorldNgayon Insight: what the first AI prescriptions change
What changed: AI prescriptions moved the licensing question from “may AI practice medicine” to “must a human sign,” and one state answered by waiving the signature — under gates. Why now: the sandbox framework had been running since 2024, the April board fight forced the oversight design, and dermatology’s scarcity provided the pilot case. Why it lasts beyond the news cycle: the agreement runs twelve months with up to two renewals, and its stage-gates, concordance thresholds and 24-hour reporting duties are the template copy every regulator and startup now studies. Who is affected: regulators deciding what oversight scales, health systems balancing access and liability, AI healthcare startups reading the entry path — and patients whose care may soon run through AI-supervised tools. What to do next: read any “AI prescribes” headline against its actual agreement — stage numbers, sampling rates, numeric gates. Risks, opportunities, costs, skills: the risk is oversight theater (sampling thinning while headlines stay loud); the opportunity is measured access where the alternative is a 61-day wait; the costs sit in review labor and monitoring; the skills gap runs through clinical-oversight engineering — the reviewers, the escalation paths and the people who build the dashboards. Connection to work and money: the same AI-governance architecture that lets a state test AI writing rules — the sovereign AI five-layer test — is now also being tested on prescribing.
For Filipino readers: the governance lesson and the health-data reality
The AI prescriptions template matters to Philippine readers for three practical reasons. First, the access parallel: the AI prescriptions debate Utah just answered dermatology scarcity is a Philippine healthcare reality too — specialist distribution is heavily Metro-Manila-weighted — and an image-based, protocol-bound AI tool is exactly the shape of telehealth pilots currently debated elsewhere in the region; Utah’s stage-gates give that debate its first working checklist (universal pre-approval, retrospective review, sampled oversight, numeric promotion gates). Second, the health-data reality: an AI prescriptions app means photo face-scans, medication histories and identities in a company’s database — which is why the privacy architecture around health data matters as much as the clinical gates; the breach post-mortems we covered in the hospital data-breach rights guide and the healthcare ransomware analysis show what happens when that layer fails. Third, the OFW angle: telehealth rules the Philippines sets will be shaped by precedents like this, and OFW families who already use telehealth across borders should watch which AI-assisted services arrive, what they state about physician oversight, and what the data-retention terms say. The governance question is universal; Utah just answered it first, with its answer published — stage numbers, gates and all.
Did Utah really allow AI prescriptions without doctors?
Partially — the accurate version: Utah’s Office of AI Policy authorized an AI system (Nolla Health’s dermatology app) to write initial topical acne prescriptions with no physician real-time signature, but the signed agreement requires two physicians to pre-approve every prescription for the first 100 patients, 100% retrospective weekly review through 500, then monthly sampling of at least 10% — plus 24-hour adverse-event reporting and state approval for each loosening stage.
What medication can the Utah AI prescribe?
Topical acne treatments only — creams and gels from a short physician-approved list — for Utah adults 18+ with mild-to-moderate acne. It cannot prescribe oral medications or isotretinoin, does not treat severe acne, and stops for pregnancy, breastfeeding, weakened immunity or a past reaction to a listed medication.
How does the Utah AI prescriptions pilot work?
Through a Regulatory Mitigation Agreement inside Utah’s AI Learning Laboratory (the 2024 sandbox law, SB 149): the AI issues prescriptions within an approved scope while licensed oversight loosens in three stages (universal physician pre-review through 100 patients; weekly retrospective review through 500; at least 10% monthly sampling after), with numeric gates (95%+ concordance, zero missed hard stops, zero serious adverse events) and the state’s written approval required before each stage.
Why is Utah the first state for AI prescriptions?
Because it built the legal vehicle first: the 2024 SB 149 sandbox law lets a state office grant time-boxed regulatory relief to test AI under safeguards — a waiver mechanism most states don’t have — and dermatology’s scarcity (the company cites a 61-day specialist wait and 11 counties without a dermatologist) provided the pilot case with the narrowest safety profile.
What did Utah’s medical board say about AI prescribing?
The Utah Medical Licensing Board pushed back hard on the earlier Doctronic AI-refill pilot: in April 2026, 11 of its 14 physician members signed a letter asking for immediate suspension over patient-safety and consultation concerns, which the state declined. For the Nolla pilot, the company met with the board before launch, and a third-party physician appointed by the board oversees prescription reviews.
Who checks the AI prescriptions before the patient gets them?
In Stage 1 (first 100 patients), two independent Utah-licensed physicians pre-approve every prescription before it is sent. In Stage 2 (500 cumulative), the AI submits directly with every case reviewed retrospectively at least weekly. In Stage 3, monthly sampling of at least 10% continues, and every adverse event is fully reviewed. Pharmacists are informed the prescription is AI-issued and given a physician to contact.
Can I use the Utah AI prescription app from another state?
No — the pilot is Utah-exclusive: patients must be Utah residents (verified), prescriptions ship only to Utah addresses, and the AI’s legal authority exists only under the state’s waiver. The company’s general app operates in other states without the AI-prescribing authority.
What happens if the AI-prescribing pilot goes wrong?
The agreement’s own controls: any serious adverse event is a numeric gate failure, adverse events must be reported to the state within 24 hours, monthly concordance reporting continues, the state may terminate the waiver at any time, and the twelve-month term requires renewal (at most twice) to continue. The pilot also cannot advance to a looser stage without the state’s prior written approval.
Sources: Utah Department of Commerce — the authorized AI pilots page (Nolla, August AI and Expect Fitness terms; commerce.utah.gov, bot-walled here); the signed Nolla Health Regulatory Mitigation Agreement PDF on the same site (bot-walled here; terms as published); IBL — the Nolla stage-gates and audit trail, read from the agreement. Utah Business (Oct 6, 2026) — the pilot’s terms and the board context; and Genetic Literacy Project (Oct 9, 2026) — the three new pilots as Healthcare Dive reported them; also the LA Times (Oct 5, 2026) report: parallaxnexus’ Bloomberg-sourced quotes (Zachary Boyd “narrow exception”; Luis Wenus on clinician-named prescriptions); Snell & Wilmer on the April 2026 board letter; Genetic Literacy Project (Oct 9, 2026) on the October 2 pilots. Company market figures (downloads, wait times, county counts) are vendor claims as announced.
