Utah lets an AI issue acne prescriptions in a closely watched pilot

Introduction

Residents of Utah can now use an app to photograph their face, have an AI system assess the severity of acne, and receive a prescription for treatment. Healthcare startup Nolla Health says its Nolla Derm service has been approved for a pilot in the state, aimed at handling relatively routine cases.

The important distinction is that the system is not merely giving a clinician a recommendation. It can autonomously issue an initial prescription. That makes the pilot a notable test of whether AI can move from supporting medical decisions to carrying out a limited clinical action on its own.

Key points

• The eligibility rules are narrow. The service is limited to Utah residents aged 18 or older who have mild-to-moderate acne. The subscription costs $4.99 per month.
• Oversight will be loosened in stages. For the first 100 patients, two physicians must approve every AI-generated prescription. During the next phase, doctors will review prescriptions after they are issued for up to 500 patients. After that, at least 10 percent of prescriptions will be sampled each month, while every escalation and side-effect case will still receive physician review.
• The treatment range is limited. Nolla Health says its system can currently prescribe eight different skin treatments. If the AI cannot confidently select a treatment, the user will be directed to a physician.
• This goes beyond renewals. Utah had already begun allowing AI systems to renew certain medications. Nolla Health says its pilot is the first in the country to issue initial prescriptions rather than simply renew existing ones.

Why it matters

Nolla Health’s stated goal is to place routine, low-acuity cases in an automated workflow so dermatologists can spend more time on complex or high-risk patients. Acne is a common condition, and an image-based service could make basic treatment easier to access for some people while reducing the need for an initial in-person visit.

The limitations are equally important. A facial image is not a complete medical examination. Lighting, camera angle, skin-tone differences, and other conditions can affect an automated assessment. Skin symptoms that look similar may also require different treatments. Once an AI can issue a prescription, the central question is no longer just whether its image analysis is accurate. It also becomes a question of who is responsible for contraindications, patient communication, adverse reactions, and incorrect decisions.

The pilot’s design creates a gradual transition from pre-approval to post-prescription review. In effect, it is testing a new division of labor: the AI handles a larger volume of routine cases, while physicians retain a role in uncertain situations, escalations, and side effects. Whether that model can expand will depend on safety records, regulatory requirements, and evidence from actual users—not simply on whether the system can recognize acne.

Utah’s experiment reflects a broader shift in healthcare AI. The next competition may not be about whether a model can interpret an image, but whether it can take action within a clearly defined clinical and legal framework. Until those boundaries are better established, AI-generated prescriptions will require closer scrutiny than an ordinary consumer health app.

Source: The Verge AI