On October 5, 2026, Nolla Health celebrated its authorization to use AI to issue initial prescriptions for acne treatment under a Utah pilot. The announcement drew enthusiastic congratulations for the promise of more affordable care, fewer barriers, and a platform that could take patients from assessment through treatment and follow-up. The pitch is access, and it is compelling: offer treatment for mild to moderate acne to people who might otherwise never see a dermatologist. Reading the announcement, though, I kept thinking about Byte.
Rember the direct-to-consumer company Byte? Byte offered a convenient, more affordable way to straighten your teeth without the usual office visits. Patients were assured that licensed dentists and orthodontists would oversee their care and that more complicated cases would be screened out. Then came the safety concerns, reports of damaged teeth, and serious questions about what that oversight actually involved.
Acne treatment and orthodontics involve different risks, and Byte’s experience does not establish that Nolla’s pilot will fail. But both models depend on getting a threshold question right: is this patient suitable for the treatment being offered? Calling the treatment “low risk” does not answer that question.
What Happened to Byte?
On October 24, 2024, Byte’s parent company, Dentsply Sirona, announced a voluntary suspension of Byte sales, marketing, and certain shipments following consultation with FDA. The following month, Byte initiated a correction that FDA classified as a Class II recall. The recall record identified a specific problem: without an in-person examination, Byte’s virtual process was inadequate to ensure that patients with conditions making treatment inappropriate were excluded. Those conditions included periodontal disease and certain serious bite and jaw problems. Patients were instructed to obtain dental evaluations to determine whether continuing treatment was appropriate. By January 2025, Dentsply Sirona had decided not to reinstate Byte’s existing at-home aligner and impression-kit offering, announcing plans for greater in-person dental oversight instead. In February, it disclosed a full write-off of the Byte trademark. Support for certain existing patients continued.
State law was also changing the economics of the business. Dentsply Sirona acknowledged that requirements for examinations and X-rays were reducing the number of prospective customers completing its process. Florida’s requirement, enacted in 2024, illustrates the point: before the initial orthodontic diagnosis and treatment, a dentist must perform an in-person examination or obtain records of one conducted within the preceding 12 months, and review appropriate diagnostic imaging. An examination requirement can look like an obstacle to access from the sales side of the business. From the clinical side, it may be how someone discovers that the patient should never have received the product. Florida did not require every subsequent encounter to occur in an office. It required a clinical foundation before treatment began.
What Actually Happened to Patients’ Teeth?
One report submitted to FDA described a patient who obtained an examination after the recall and learned that their bite condition required surgery, making Byte treatment inappropriate. That does not establish that the aligners caused the surgical problem. It raises the more basic question of why the patient was receiving that treatment at all. Another patient report described front teeth colliding while back teeth stopped making contact, accompanied by pain and a worsening bite. The patient alleged that additional aligners and advice to continue treatment made matters worse. A report received in April 2026 described loose and chipped teeth and unsuccessful attempts to obtain clinical assistance. These reports are not FDA findings that Byte caused each injury.
The May 2025 amended securities complaint against Dentsply Sirona alleged that reports involving at least 115 instances of tooth loss, 110 instances of pulp necrosis, and 1,160 tooth fractures were not timely reported to FDA. Those figures are allegations, not established totals of distinct injured patients. The complaint also described an alleged breakdown in clinical oversight: dentists could approve treatment plans with four clicks, sometimes in under two minutes; rejected plans could be reassigned to other dentists; and patients seeking help were handled by support personnel rather than given meaningful access to their prescribing dentists.
In a January 16, 2026 decision, the court allowed central securities claims concerning oversight, screening, and FDA reporting to proceed. Although this was a ruling on the allegations rather than a finding after trial, the court rejected the argument that an initial prescription necessarily fulfilled representations of continuing, doctor-directed care.
Healthcare companies should pay attention to that reasoning. When a company promises physician oversight, what the physician actually does becomes part of the legal analysis. A prescription bearing a clinician’s name does not, by itself, establish that the patient received the continuing care being advertised.
So What, Exactly, Has Utah Allowed Nolla to Do?
Under the signed agreement, Utah’s Office of Artificial Intelligence Policy and Division of Professional Licensing provide limited relief from enforcement of specified professional practice and telehealth requirements. The relief applies to the activities authorized by the agreement, which expressly states that participation does not constitute state endorsement or approval of the technology. The pilot covers “eligible” Utah adults with mild to moderate acne and specified topical treatments. It excludes oral medications, including isotretinoin, and patients with specified conditions, including pregnancy, planned pregnancy, and breastfeeding. Those exclusions help define the risk being tested.
Physician review changes in stages:
- Initially: Two physicians independently review and approve every prescription for the first 100 patients before issuance.
- Next: Every case receives retrospective physician review, at least weekly, through the next 400 patients.
- Later: Physicians review at least 10% of prescriptions monthly, plus every case involving an escalation or side effect.
Advancement requires minimum testing periods, defined safety measures, including at least 95% agreement with physicians and zero serious adverse events, and written state approval. Reaching the patient count does not automatically authorize less review. These are substantive safeguards, but their function changes as the pilot advances. Reviewing a prescription before issuance can prevent an error from reaching the patient. Retrospective review may identify it only after treatment has begun. And agreement between an AI system and a physician measures something different from the patient’s outcome over time. The evaluation needs to account for both.
And What Has FDA Allowed?
Utah’s authority has limits. It cannot waive federal medical device requirements, and its own FAQ expressly acknowledges that the agreements do not override FDA jurisdiction. The public materials reviewed for this article do not establish Nolla-specific FDA clearance, approval, or an enforcement discretion determination. That does not establish noncompliance. It means federal regulatory status remains a separate question that cannot be answered by pointing to Utah’s agreement.
Under FDA’s January 2026 clinical decision support guidance, whether software qualifies for exclusion from device regulation depends on its functions. Relevant questions include whether it analyzes medical images and whether a healthcare professional can independently review the basis for its recommendations. FDA specifically includes dermatology images in its discussion of medical images. An app that analyzes a facial scan and selects treatment therefore requires more analysis than simply labeling it clinical decision support.
Note, however, Byte itself had an FDA-cleared prescription aligner system. That clearance did not resolve whether its screening, supervision, or response to patient complaints was adequate. Federal device regulation and state regulation of clinical practice address different questions, and both can matter to the same service.
Whose Medical License Is on the Prescription?
My initial reaction to Nolla’s announcement included a corporate practice of medicine question: if AI makes the diagnosis and selects the prescription, who is practicing medicine? Utah law permits physicians to practice through various business entities, while reserving the practice of medicine within those entities to appropriately licensed individuals. The company’s involvement alone therefore does not establish an unlawful arrangement.
According to Utah’s explanation, every prescription issued through an authorized pilot is transmitted under the NPI and license of an overseeing physician who serves as prescriber of record. That identifies a human practitioner, but it leaves important questions about the practitioner’s role:
- What did the physician review?
- Who can override the AI, stop prescribing, or require an examination?
- What happens when the physician and the company disagree?
If a patient is harmed, the analysis could involve the physician’s conduct, the company’s design and operation of the system, and failures in screening or follow-up. Limited relief from state enforcement does not decide in advance who would be liable for those failures.
Utah also requires that Nolla’s patient agreements not disclaim or limit its liability for physical, medical, or legal harm resulting from the authorized AI outputs. Patients should not discover that the company’s confidence in its technology disappears somewhere inside the terms of service.
What Happens When the Patient Needs More Than the App Can Provide?
Byte’s experience also makes follow-up difficult to dismiss as a secondary concern. Treatment can be appropriate when prescribed and still require a clinician’s attention later. Nolla’s protocol provides direct physician access at no additional cost, escalation procedures, and a typical physician response within 24 hours. Those are useful commitments. Access to messaging, however, should not be confused with an immediate clinical response.
The practical test is what happens after a concern arises. How quickly is an urgent complaint recognized and acted upon? Does a referral result in an actual appointment? If conventional care was unaffordable or unavailable when the patient entered the program, how is that barrier addressed when the app can no longer manage the problem? Directing someone to in-person care may be exactly the right clinical decision. An access model still needs to account for whether that patient can obtain it. Otherwise, the advertised price tells us what it costs to start treatment, while the cost of complications and unfinished care falls elsewhere.
What Should Healthcare Organizations and AI Vendors Take From This?
Before relying on promises about access and physician oversight, examine how the service will operate when a patient is unsuitable, a recommendation is wrong, or treatment does not go as expected:
- Test the screening process. Identify what information the system needs, what it cannot reliably assess remotely, and when missing information stops treatment. Review inappropriate approvals as closely as successful cases.
- Define the physician’s actual work. Specify what is reviewed, when review occurs, who can override the system, and how quickly patients reach a clinician. Public statements about oversight should reflect those duties.
- Make follow-up and referrals workable. Establish who handles escalation, arranges further care, and confirms the patient received it. Include continuity arrangements if the service closes.
- Document regulatory authority and financial responsibility. Identify the state relief and federal analysis supporting each function. Address insurance, indemnification, and responsibility for failures across the company, clinicians, and other vendors.
Better access to affordable care is a goal worth pursuing, and Nolla’s pilot offers a chance to test what AI can contribute. Its results should be assessed through patients’ outcomes, including whether those who need additional attention are identified and connected with a clinician promptly.
Byte’s experience provides useful lessons about patient screening and what effective oversight requires in practice. Applying those lessons early could help Nolla and similar programs deliver the convenience patients want with the continuing care they may need.