Utah barred mental-health chatbots from selling chat data

What happened

Utah's House Bill 452, the 'Artificial Intelligence Amendments' act, took effect on 7 May 2025. The enrolled bill enacts a new chapter of the Utah Code, Chapter 72a, 'Artificial Intelligence Applications Relating to Mental Health', built around a single defined product: a mental health chatbot. The statute defines that term as an artificial intelligence technology that uses generative AI to hold interactive conversations similar to the confidential communications a person would have with a licensed mental health therapist, where the supplier represents, or a reasonable person would believe, that it can provide mental health therapy or help manage or treat a mental health condition. It excludes scripted tools such as guided meditations and systems that only route a user to a human therapist.

What the documents show

The enrolled text bars a supplier from selling or sharing a Utah user's identifiable health information or raw input with any third party, subject to narrow exceptions such as a user-directed request to their own health plan. It bars using a mental health chatbot to advertise a product inside the conversation unless the advertisement is clearly disclosed as such, and bars using a user's input to target or customise advertising to that user. It requires the chatbot to disclose, before use and repeatedly, that it is artificial intelligence and not a human. Utah's Division of Consumer Protection administers the chapter, with fines up to $2,500 per administrative violation and civil penalties up to $5,000 per court-ordered violation.

The mechanism

The statute's reach is defined by function, not by branding: any product that behaves like therapy, simulating the confidential exchange a licensed therapist would hold, falls inside the definition regardless of what the company calls it, while a wellness app that only plays a recorded meditation does not. A separate section, added to the state's professional-licensing title, gives a supplier an affirmative defence to certain licensing claims if it files a written safety policy with the division and follows it, covering clinician involvement, adverse-outcome monitoring and real-time risk protocols; the defence does not cover a full range of consumer claims and does not make the chatbot a licensed therapist.

What it leaves open

Whether a general-purpose companion app that sometimes discusses feelings, without being represented as therapy, meets the 'reasonable person would believe' branch of the definition is a judgment call left to enforcement and litigation. The filed-policy affirmative defence is optional, so its existence does not tell a user whether a supplier has actually adopted it. This is an editorial reading: a companion app marketed for emotional support sits close to this law's boundary without necessarily crossing it.

• Does the app or its marketing represent that it can provide therapy, or only that it can talk about feelings?
• Has the supplier filed a safety policy with Utah's Division of Consumer Protection, and does the app say so?
• Is the user's input being used anywhere to select or personalise advertising?

By defining the regulated product around a therapeutic function rather than a technology or a company, Utah's act draws a boundary a growing companion-app market will keep testing from both sides.