Illinois, Nevada, Rhode Island and Maine have now drawn a hard legal line that didn’t exist a year ago: an AI chatbot cannot practice therapy on a human being. Utah has taken a different path, allowing the bots to keep talking as long as they admit what they are. Together, the two approaches mark 2026 as the year state legislatures stopped watching the AI-therapy boom from the sidelines and started writing enforceable law around it.
Illinois Draws First Blood
Illinois moved first. Governor JB Pritzker signed the Wellness and Oversight for Psychological Resources Act, known as WOPR, on August 1, 2025, and it has been reshaping how behavioral health companies operate in the state through 2026. The law, based on House Bill 1806, prohibits any person or company from offering therapy or psychotherapy services through AI unless a licensed professional is directly performing the work. Even licensed therapists can’t hand an AI system the wheel: the law bars using AI to make independent therapeutic decisions, to directly conduct therapeutic communication with a client, or to generate treatment recommendations without a human clinician’s review and sign-off. Violations carry civil penalties of up to $10,000 per infraction, enforced by the Illinois Department of Financial and Professional Regulation.
What’s Still Allowed
The Illinois law isn’t a blanket ban on AI touching mental health care. Administrative and supplementary uses remain legal: AI can manage scheduling, handle billing, and, with a client’s written consent, draft session notes, run ambient documentation during appointments, analyze anonymized data, and flag referrals. The distinction lawmakers drew is between AI as an assistant to a licensed human and AI as the therapist itself. The bill passed the Illinois General Assembly nearly unanimously, driven in part by news reports of chatbots giving users harmful or dangerously inaccurate mental-health advice.
Nevada, Utah and a Widening Map
Nevada followed with AB 406, effective July 1, 2025, which similarly prohibits companies from offering AI systems that provide professional mental or behavioral healthcare, and bars licensed clinicians from using AI to deliver therapy directly. Rhode Island and Maine have since joined Illinois and Nevada in banning AI from delivering therapy to the public outright, according to tracking by industry publication Becker’s Behavioral Health, bringing the total to four states with outright bans as of mid-2026.
Utah chose a lighter touch. Its HB 452, rather than banning AI chatbots, requires them to clearly disclose that users are talking to software — at first contact, again if a user returns after a week away, and any time the user asks. The law also bars mental-health chatbot operators from selling or sharing identifiable mental-health data and creates a rebuttable legal presumption of compliance for companies that follow the disclosure rules. That model of letting AI operate but forcing it to be honest about what it is is spreading: California, Colorado and Maine have all enacted or amended chatbot-disclosure requirements in 2026, and multistate policy trackers describe three distinct regulatory camps emerging nationally — outright bans, disclosure-first frameworks, and crisis-detection mandates paired with private rights to sue.
Why Now
The regulatory wave follows a string of publicized incidents involving AI companion and therapy-style apps giving users advice that mental health professionals characterized as inappropriate or dangerous, particularly for minors and people in crisis. Lawmakers in both parties have cited the absence of any licensing, training or liability standard for AI systems marketed as emotional support tools, in contrast to the years of supervised clinical training required of human therapists.
The Industry Pushback
AI mental-health app makers argue the bans risk cutting off free or low-cost support for the roughly half of Americans with a diagnosable mental health condition who currently get no treatment at all, citing long waitlists and therapist shortages in rural areas. Some companies have restructured their products specifically to comply with Illinois and Nevada’s rules, repositioning chatbots as journaling or coping-skill tools rather than therapy, while others have simply geofenced their apps to exclude those states. Critics of the ban model counter that Utah-style disclosure requirements do little to stop a chatbot from giving harmful advice — they just require it to admit, mid-crisis, that it’s a machine.
What’s Next
With four states now banning AI therapy outright and disclosure rules spreading to California, Colorado and Maine, expect the map to keep filling in through the rest of 2026 as more legislatures convene. Federal regulators, including the FDA and FTC, have so far left this almost entirely to the states, and unless Congress acts, the next year will likely produce a genuinely fractured national landscape — one where the legality of an AI therapy chatbot depends entirely on which state line a user happens to be sitting behind.
Photo: Matheus Bertelli / PEXELS via Pexels
