Utah’s Office of AI Policy outlines sandbox activity, mental‑health and youth risks, and national coordination

The interim Business, Labor and Commerce Committee received the 2025 annual report for the Office of AI Policy from Director Zach Boyd, who described three office programs — a regulatory mitigation (sandbox) program, a learning laboratory for deep policy studies, and continuous observation of AI‑related risks — and discussed recent activities and future bills.

‘‘We have three tasks that you guys have given us,’’ Boyd said. ‘‘First is to balance protection of the public with fostering of innovation.’’ Boyd told the committee the office aims to provide temporary regulatory relief for carefully scoped AI pilots, to gather outcome data, and to recommend permanent legal changes when practices and best practices become clear.

Boyd said the AI sandbox functions similarly to the GOEO sandbox: companies apply and the office crafts a temporary, written agreement that defines testing conditions, required data collection and reporting. He said most initial inquiries do not require mitigation, but that applications have accelerated: ‘‘in the last 3 or 4 months, things have really had an uptick. We're getting about one application a week now from companies in all kinds of sectors,’’ Boyd said.

Examples presented to the committee included a public‑health dentistry pilot in which computer‑vision models help hygienists identify a narrowly defined set of common, low‑risk conditions (periodontal disease) from X‑rays and visual inspection, enabling a lower‑cost service model for homeless and low‑income populations. Boyd said academic literature and the dental board supported a narrow pilot that included outcome reporting; early results, he said, ‘‘are indeed very good.’’

Boyd also described engagement with a mental‑health chatbot company (referred to in the presentation as Eliza Chat). The office negotiated governance measures rather than a statutory waiver and agreed in one case to allow a 30‑day cure period for the company to fix issues if the bot exceeded its intended boundaries. Boyd said that approach gives startups a predictable path to test responsibly.

The learning laboratory has studied AI in mental health and recommended a mix of consumer protections, a safe harbor to encourage compliant firms to enter the market, and education for clinicians. The legislature adopted the office’s mental‑health chatbot recommendations in a bill last session, Boyd said, and he told the committee his office has begun new studies on deepfakes and ‘‘AI companions’’ (personal AI chatbots used for social or intimate interaction). Boyd said early evidence suggests a significant share of teens and emerging adults — ‘‘somewhere between a fifth and a third’’ — have used companion‑style AI tools, and he described harms that range from worsening social isolation to rare tragic outcomes in individual cases.

Director Boyd and Executive Director Margaret Busse of the Department of Commerce also described federal coordination. Busse said an early congressional draft would have imposed a state moratorium on AI regulation or conditioned federal grant money on states not regulating AI; state officials and lawmakers successfully opposed that language. Busse said the administration’s AI action plan now contemplates a federal sandbox to coordinate with state sandboxes, and Boyd said coordination would allow firms to work concurrently with state and federal regulators.

Committee members asked about agency and municipal limits; Boyd and Busse clarified the office can only affect state statutes and administrative rules, not federal law or municipal ordinances. Members also raised liability, scope‑of‑practice and government use of AI in state agencies. Boyd said liability and contracts are unresolved industry concerns and that some hospitals are worried vendor contracts place AI liability on providers even after hospital‑level procurement decisions.

On youth and consumer harms, Boyd described the industry dynamics that can incentivize high engagement. ‘‘The same companies to some degree... are all building these kinds of AI companions,’’ said Busse. The office indicated it plans to propose legislation informed by the learning lab’s studies on deepfakes and companion apps.

Why it matters: The Office of AI Policy is both a regulator and a convenor: it negotiates temporary relief for pilots, studies systemic risks through its learning lab, and advises the legislature on bills. The office’s mental health recommendations were already adopted; the committee heard that deepfake and companion‑app bills are under development.

What’s next: Expect more committee briefings and potential bill files in the next session addressing deepfakes, companion apps, consumer privacy and coordination of state and federal sandboxes.

Speakers quoted in this article are identified from the committee meeting transcript and appeared during the Office of AI Policy presentation and Q&A.