Writing in npj Digital Medicine in 2026, Nina de Lacy of the University of Utah and Zachary Boyd of the state's Office of Artificial Intelligence Policy describe how Utah has set about regulating mental-health AI, and argue the approach could serve as a template for others. The starting point is access: half of US residents live in areas with a shortage of mental-health professionals, and many people with anxiety or depression go untreated, which is part of what draws them to chatbots.

Rather than banning the tools or leaving them unregulated, Utah's framework creates a "safe harbour" for mental-health AI that meets defined standards. These include pre-deployment safety testing, protocols for escalating a user in crisis to a human, clinical oversight, ongoing monitoring, and consumer protections covering data privacy and advertising. The authors characterise this as risk-benefit regulation rather than risk-only regulation.

The analysis also sets out how the rules were drawn up. A multi-stakeholder review deliberately included people with lived experience of mental illness, among them those in care facilities and the parents of affected children, alongside clinicians, technologists, and academics. The authors present the exercise as a worked example for other jurisdictions weighing how to govern a class of product already in consumers' hands.

Whether a safe-harbour regime improves outcomes, rather than simply formalising a market, is an empirical question the framework does not yet settle. As a piece of governance design, though, it offers a concrete alternative to the binary of prohibition or laissez-faire.