Calm AI Therapy Editorial · 8 min read

Several US States Restricted AI Therapy. Here Is What Each One Actually Says.

Illinois, Nevada and Utah took three different approaches to regulating AI mental health tools, and California and New York took a fourth. A plain-language guide to what changed and what it means for you.

Published September 16, 2026

In 2025 and 2026 a number of US states wrote laws about AI and mental health, and they did not agree with each other. If you use one of these products, or build one, the differences matter more than the headlines do.

Illinois went furthest. The Wellness and Oversight for Psychological Resources Act, in force since August 2025, prohibits providing, offering or advertising therapy or psychotherapy through AI unless a licensed professional is in charge of the service. The advertising clause is the part people miss: marketing an app as an AI therapist is itself treated as unlawful, with penalties running to five figures per violation. General wellness apps are exempt, which is the category most of these products actually fall into.

Nevada took a similar line in Assembly Bill 406, barring AI from standing in for a counsellor or psychologist, and reaching into schools specifically.

Utah went the other way. House Bill 452 did not ban anything. It requires a mental health chatbot to say plainly that it is software, at first contact, again when someone returns after a break, and any time they ask. It restricts what the product may advertise and what it may do with the data people hand over. Notably it offers a safe harbour: a provider that files a written compliance policy gets an affirmative defence against liability. That is the only constructive path any state has offered so far.

California and New York took a fourth approach again, requiring crisis detection and backing it with a private right of action, which means an individual can sue rather than waiting for a regulator.

The practical effect is a patchwork, and the honest summary is that the same product can be lawful in one state and not in the one next door. If you are using one of these tools, the thing worth knowing is not which law applies to you, it is what the tool does. Does it tell you it is software without being asked? Does it claim to treat anything? Does it know what to do when the conversation turns serious?

We build to the strictest rule we know of rather than maintaining several versions of the truth. Aura tells you she is software without being asked. She does not present herself as a clinician, because she is not one. Every conversation passes through a crisis layer that routes to the emergency service for your country rather than trying to handle it alone.

None of that is a legal opinion, and this article is not one either. It is a description of a landscape that changed quickly and is still moving. If you are somewhere with a restriction, take it seriously: the laws exist because products in this category caused harm, not because legislators were bored.