What happened
Utah’s Office of Artificial Intelligence Policy (OAIP) unveiled three new health‑care pilots in its AI sandbox: an AI system that can refill chronic‑condition prescriptions, a tool that writes initial acne prescriptions, and a virtual physical‑therapy platform. The pilots follow earlier tests that let Doctronic’s AI refill certain prescriptions under clinician oversight. After pushback from the Utah Medical Licensing Board, the state removed two drugs—Butalbital and Flecainide—from the AI‑refill list. Regulators are still debating how liability will be split between AI vendors and supervising physicians; for now, sandbox participants must carry special liability insurance and cannot waive liability. OAIP director Zach Boyd said a liability‑focused bill is expected in the coming months. The sandbox also faces uncertainty about FDA jurisdiction, as the agency may require pre‑market authorization for the AI tools. Six independent evaluators—including the Coalition for Health AI and Stanford Clinical Excellence Research Center—have been contracted to assess technical claims and ongoing performance. Utah has master agreements with Intermountain Health and University of Utah Health to source future AI pilots.
At the ADAPT conference in Santa Fe, OAIP director Zach Boyd announced three new health‑care pilots under Utah’s AI sandbox: a chronic‑condition prescription‑refill system, an AI that drafts initial acne prescriptions, and a virtual physical‑therapy solution. The sandbox, launched in 2024, offers participating companies regulatory relief in exchange for state oversight.
Following criticism from the Utah Medical Licensing Board, the state removed two drugs—headache reliever Butalbital and antiarrhythmic Flecainide—from the AI‑refill list. Regulators continue to debate liability allocation; currently, sandbox participants must maintain special liability insurance and cannot waive responsibility for AI‑driven actions.
Utah is drafting a liability‑focused bill expected to be introduced within the next two months, aiming to create a multi‑year framework for assigning legal responsibility. Simultaneously, the state is engaging with the FDA to determine whether its sandbox AI tools will require pre‑market authorization.
Six independent evaluators—Coalition for Health AI, Clarion AI Partners, Glacis Technologies, mpathic AI, Vega Health, and Stanford Clinical Excellence Research Center—have been contracted to assess technical claims, data handling, and ongoing model performance. Evaluators will provide continuous assessments as AI models evolve.
Master agreements with Intermountain Health and University of Utah Health allow these systems to propose future AI pilots, expanding the sandbox’s reach within the state’s health‑care ecosystem.
Why it matters
The expansion of Utah’s AI sandbox into prescription refills, acne treatment and virtual therapy puts a state‑level regulatory experiment at the forefront of U.S. health‑care AI deployment. How Utah resolves liability—whether physicians or vendors bear responsibility—could set a template for other jurisdictions grappling with AI‑induced medical errors. The state’s dialogue with the FDA may shape a unified regulatory pathway, influencing how AI‑driven diagnostics and treatment tools are cleared nationally. Moreover, the requirement for independent technical evaluators introduces a transparency mechanism that could become a model for ensuring safety and efficacy as AI models evolve over time. The pilots also test the feasibility of granting regulatory relief in exchange for oversight, a balance that could accelerate innovation while protecting patients.
The pilots test a regulatory approach that balances innovation speed with patient safety, offering a potential blueprint for other states and federal agencies.
Liability decisions will influence how health‑care providers and AI vendors negotiate risk, potentially affecting insurance markets and the willingness of companies to enter the sandbox.
Clarifying FDA jurisdiction could either streamline AI deployment through a unified pathway or impose additional compliance burdens, impacting the pace of AI adoption in clinical settings.
Independent technical evaluation introduces a layer of accountability that may become a standard requirement for AI‑driven medical tools, ensuring that model updates do not erode safety or efficacy.
Successful integration with major health systems could demonstrate scalability, encouraging broader adoption of sandbox‑style oversight across the United States.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Why can ethical evaluation not be reduced to one model score?
What to watch next
Watch for the introduction of the liability bill during Utah’s interim legislative session and any subsequent amendments that define vendor versus clinician responsibility. Monitor FDA statements or guidance that clarify whether sandbox AI tools will need pre‑market clearance. Follow the first‑round evaluation reports from the six independent assessors to see how technical claims are validated and whether the sandbox’s oversight model proves scalable. Finally, track whether additional health systems join the sandbox and if the pilots expand beyond the initial drug list and therapeutic areas.
Passage and content of Utah’s upcoming liability bill, especially provisions that assign responsibility to AI vendors versus clinicians.
FDA guidance or rulings on whether sandbox AI tools fall under existing medical device regulations.
First evaluation reports from the six independent assessors, focusing on model performance, mitigation, and data security.
Expansion of the sandbox to additional health‑care providers or therapeutic areas, indicating the model’s sustainability.
Reactions from the Utah Medical Licensing Board and other stakeholder groups as the pilots progress.