Mental‑health AI regulation heats up

Artificial intelligence is moving deeper into mental-health care, and the argument has shifted. The live question is no longer whether people will use chatbots for therapy, triage, or medication management. They already do. The question is who will set the rules before those systems start acting like clinicians in everything but name. Utah has emerged as the first state seriously trying to answer that question, with a new body of law for mental-health chatbots and a regulatory sandbox that is now letting one startup automate parts of psychiatric prescribing (nature.com) (le.utah.gov) (theverge.com). That did not happen in a vacuum. Utah’s 2025 law, HB 452, created disclosure, privacy, and advertising rules aimed specifically at mental-health chatbots used by people in the state. Separate Utah AI laws had already created an Office of Artificial Intelligence Policy and a regulatory sandbox for testing higher-risk systems under state supervision. In a commentary published March 27 in *npj Digital Medicine*, researchers involved in that process said Utah’s early review helped shape both the legislation and guidance for therapists using AI tools. Their basic point was blunt: mental-health chatbots are already in people’s pockets, so the state should regulate the real product that exists, not the hypothetical one lawmakers wish they were debating (le.utah.gov) (regulations.ai) (nature.com). Now the sandbox is testing the hardest case. Utah has approved a one-year pilot for Legion Health that allows its AI system to renew some psychiatric prescriptions without a doctor signing each refill. Reporting on the approval says the pilot is limited. The system is for stable patients, not new diagnoses. It covers a narrow set of medications, including fluoxetine and sertraline. Utah’s Office of Artificial Intelligence Policy approved the program, and the company sells the refill service as a low-cost subscription. That is a remarkable sentence to be able to write in April 2026: in one state, a chatbot can now do part of the work that used to require a prescriber’s direct signoff (msn.com) (theverge.com) (futurism.com). That is exactly why clinicians are getting more explicit about what these systems are and are not. A JAMA Psychiatry article published online April 1 argued that therapists and psychiatrists should routinely ask patients how they are using AI for emotional support, advice, and mental-health information. The authors pointed to evidence that more than 5 million U.S. youths have already sought mental-health advice from AI, and that among adults with mental-health conditions who use large language models, nearly half report using them for support. The authors’ point was not that AI use is rare and alarming. It was that it is common enough to belong in a standard clinical history, alongside sleep, alcohol use, and medication adherence (jamanetwork.com) (boisestatepublicradio.org). That recommendation lands at the same moment health systems are pushing AI into the workforce itself. NPR reported on April 7 that mental-health providers are already using AI tools for note-taking and recordkeeping, while some workers fear that scripted intake and triage systems are becoming a bridge to further automation. Kaiser Permanente told NPR that its use of AI does not replace clinical expertise and that it is only evaluating, not yet deploying, a tool from the U.K. company Limbic to help members access care. Even so, the labor fight NPR described makes the shape of the next conflict obvious. The first wave of concern was about patients talking to bots. The next wave is about institutions deciding which parts of care can be turned into software (npr.org) (kedm.org). Utah’s answer, for now, is not to ban the software and not to trust it. It is to narrow the use case, force disclosure, watch the pilot, and keep humans legally accountable around the edges. That may prove too permissive. It may prove more realistic than the rest of the country’s drift. But it is at least a real regulatory theory, and right now that makes Utah unusual. Most places are still arguing over whether AI belongs in mental health. Utah is already deciding whether a stable patient can get a refill for Prozac or Zoloft from a chatbot (nature.com) (biztechweekly.com) (theverge.com).