Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It

A $4.99 App Now Prescribes Acne Drugs in Utah With No Doctor in the Room — and the Oversight Ladder After 500 Patients Is the Real Story

A $4.99 App Now Prescribes Acne Drugs in Utah With No Doctor in the Room — and the Oversight Ladder After 500 Patients Is the Real Story

Nolla Health's AI can now scan a user's face, grade acne severity and autonomously write a prescription in Utah — the first US program to issue initial prescriptions rather than renewals, at $4.99 a month. The oversight is deliberately gradual: two physicians check every script for the first 100 patients, then review after the fact for up to 500, then only a sample of at least 10 percent a month.

People in Utah can now get a prescription for acne treatment written by an AI system that never examines them in person. On Monday, healthcare startup Nolla Health announced that residents of the state can scan their face with its app, let the AI grade the severity of their acne, and receive a prescription autonomously — a program Bloomberg first reported and The Verge confirmed with the company. The service costs $4.99 a month, is limited to Utah residents aged 18 and over with mild-to-moderate acne, and can currently prescribe eight different skin treatments.

What makes this a first is the word "initial." Several companies have been experimenting with AI-generated prescriptions in Utah, which began allowing AI systems to renew certain medications earlier this year: August AI was approved to pilot AI-assisted renewal of prescriptions for chronic conditions, and Doctronic has run an AI prescription refill pilot in the state. Nolla says it is the first in the country to have an AI issue the original prescription itself rather than refilling one a human doctor once wrote. That distinction matters — a renewal inherits a human's clinical judgment, an initial prescription replaces it.

The safety architecture is a phased loosening of physician oversight, and the company has published the thresholds. For the first 100 patients, two physicians approve each AI-generated prescription before it is issued. For the next phase, covering up to 500 patients, physicians review prescriptions only after they have already been dispensed. After that, the company says physicians review a sample of at least 10 percent of prescriptions every month, plus every case involving an escalation or a side effect. If the AI cannot "confidently select a treatment," the user is directed to a physician.

Read those numbers as an engineering document and the design is coherent: front-load human review where the model is least proven, then step it down as real-world evidence accumulates, and always audit escalations and adverse events. Read them as a patient and the arithmetic is harder to love. In steady state, roughly nine out of ten AI-prescribed scripts will face no contemporaneous human review at all — the 10 percent sample is retrospective, and the "every escalation or side effect" catch only triggers if something goes visibly wrong and gets reported. Automated systems fail in correlated, systematic ways; a monthly sample is a reasonable quality-control tool, but it is not a diagnosis of what the model is systematically missing.

The company's own framing concedes the trade. "The pilot is designed to complement, not replace, physicians," Nolla writes, arguing that "by handling routine, low-acuity cases, Nolla Derm frees dermatologists to focus on complex, high-risk patients." There is a real access case here — acne is one of the most common reasons people see a dermatologist, treatment is protocolized, and a $4.99 app is cheaper and faster than a referral queue. The risk is not that an AI handles mild acne badly in the average case; it is that the population that gets filtered out — the person whose "mild" acne is actually something else — depends on an AI confidence threshold nobody outside the company can inspect.

The broader read: Utah has effectively turned itself into the national test bench for autonomous prescribing, first with renewals and now with initial prescriptions, and every other state's medical board is watching the same experiment. The precedent being set is not that AI can write prescriptions — August AI and Doctronic already moved that line for renewals — but that the unit of regulatory proof is becoming a phase-gated pilot with published oversight thresholds rather than a clinical trial. Whether that is a faster, smarter path to safe deployment or a quieter way to normalize skipped review depends entirely on whether the escalation data ever gets published. Nolla should be pushed to make it public.

Comments (0)

Log in to join the discussion

Log In

No comments yet