A JAMA Forum commentary by Michelle M. Mello, JD, PhD, MPhil, published in February 2026, analyzing a Utah pilot program begun in January 2026 that allows Doctronic, an AI developer, to deploy an autonomous AI agent that renews prescriptions for consumers who request it. Under the program, Utah's Division of Professional Licensing agreed not to enforce its unprofessional-conduct laws against Doctronic, contingent on the developer adhering to a contract with safety and privacy protections. The arrangement is the first US state-level regulatory-experimentation regime described in the source as explicitly allowing an autonomous AI to make consequential medical decisions at scale.
Author: Michelle M. Mello, JD, PhD, MPhil Publication: JAMA Forum (Journal of the American Medical Association) Released: February 2026
Pilot design
The pilot covers 192 drugs for chronic conditions, including hypertension, depression, birth control, diabetes, glaucoma, asthma, inflammation, and hyperlipidemia, and is limited to adult patients. It proceeds in three phases. In the first phase, Doctronic-employed physicians review the AI agent's output before any action is taken, for the first 250 patients. In the second phase, the system acts autonomously and Doctronic conducts retrospective review of the next 1,000 patients. In the third phase, the agent moves to full autonomous operation if the earlier phases pass.
The regulatory mechanism is conditional non-enforcement of state unprofessional-conduct laws rather than category-based rules. Additional safeguards take the form of contract-based safety and privacy protections; the specific terms are not given in the JAMA Forum text.
| Element | Detail |
|---|---|
| Eligible drugs | 192 drugs for chronic conditions including hypertension, depression, birth control, diabetes, glaucoma, asthma, inflammation, hyperlipidemia |
| Eligible patients | Adults only |
| Phase 1 | Doctronic-employed physicians review the AI's output before action for the first 250 patients |
| Phase 2 | Retrospective review of the next 1,000 patients after the system has acted |
| Phase 3 | Full autonomous operation if pilot phases pass |
| Regulatory tool | Conditional non-enforcement of state unprofessional-conduct laws |
| Other safeguards | Contract-based safety and privacy protections (specific terms not in the JAMA Forum text) |
Mello's analysis
Mello frames the pilot in structural terms and neither dismisses nor endorses it. She argues that the underlying problems the pilot addresses are real: many patients face cost or access barriers to seeing physicians for renewals, particularly given mounting rural clinician shortages and rollbacks of Medicaid coverage and ACA subsidies. Renewals, she writes, are unreimbursed work added to "an already crushing administrative load," and many chronic-condition prescriptions change little over time. On this account the pilot deserves analysis rather than dismissal, even if the chosen solution is contestable.
Mello reports that reaction from professional associations has been negative, with physicians' and pharmacists' associations arguing that AI "should NOT be making care decisions." Based on her review of Doctronic's proposal and contract, she characterizes the pilot as well-designed, with several guardrails against patient harm.
The commentary does not take a final position on whether the pilot will succeed. It is an early-stage policy analysis of a deployment expected to produce empirical evidence over the following year, and Mello notes that the program may succeed, fail, or be amended significantly as that evidence accumulates.
Provenance
Mello's account of the pilot design is grounded in her review of Doctronic's proposal and contract; the implications she draws for the future trajectory of such regimes are more provisional, given that the pilot is still in progress.
The regulatory mechanism Mello describes, conditional non-enforcement of state professional-conduct laws, generalizes beyond pharmaceuticals: she notes it could in principle be applied to any state-licensed profession, such as legal services, accounting, real estate, or teaching. The pilot is the first documented state-level regulatory-experimentation regime for autonomous AI clinical decision-making at scale, and contrasts with category-based state approaches: Colorado's AI Act regulates by category (high-risk AI), while Utah regulates by negotiated, case-by-case contract.
Mello is also a co-author of the Stanford HAI policy brief on AI in health insurance (Toward Responsible AI in Health Insurance Decision-Making — Mello, Trotsyuk, Djiberou Mahamadou, Char (Stanford HAI Policy Brief, February 2026)), which applies a similar structural-problem framing to a different healthcare domain.
Relationships
- supports: Healthcare — AI Deployment (state-level autonomous-AI clinical deployment), Agentic AI (state-sanctioned deployment), Regulatory Experimentation (planned)
- related: Toward Responsible AI in Health Insurance Decision-Making — Mello, Trotsyuk, Djiberou Mahamadou, Char (Stanford HAI Policy Brief, February 2026) (Mello as co-author; cross-domain healthcare-AI lens), Michelle Mello (planned), FDA — Food and Drug Administration (AI Deployer) (regulatory-gap context), Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment) / California AB 3030 (Healthcare AI Disclosure) (alternative state-level regulatory philosophies)
- instance-of: Responsible AI Deployment