AI Therapy Chatbot Bans: Where States Draw the Line

As of September 2026, at least seven US states restrict AI from acting as a therapist: Illinois, Nevada, Maine, Vermont, Rhode Island, Colorado and Tennessee. A second group, led by Utah, California and New York, lets mental-health and companion chatbots operate but regulates how they behave. In the first group, the line is not the technology itself. It is who delivers the therapy and what the product claims to be.

None of these laws makes it illegal to type your worries into a general-purpose chatbot. They make it legally risky to sell AI as therapy, or for a clinician to let AI talk to clients. The same app can be lawful in Salt Lake City and an unfair trade practice in Portland, Maine. This guide maps the patchwork from the bill texts themselves.

Which states ban AI therapy outright?

Illinois set the template. Governor JB Pritzker signed HB 1806, the Wellness and Oversight for Psychological Resources Act, on August 1, 2025, and it took effect immediately. The core clause says an individual, corporation or entity may not "provide, advertise, or otherwise offer therapy or psychotherapy services, including through the use of Internet-based artificial intelligence," unless a licensed professional conducts them. Licensed clinicians may use AI only for administrative support (scheduling, billing) and supplementary support (notes, referrals). They may not let it make independent therapeutic decisions, interact directly with clients in therapeutic communication, produce treatment plans without their review, or "detect emotions or mental states." The Illinois Department of Financial and Professional Regulation (IDFPR) enforces the act, with civil penalties of up to $10,000 per violation. According to the IDFPR announcement, the bill passed both chambers unanimously.

Nevada took a product-side approach. Assembly Bill 406, effective July 1, 2025, bars an AI provider from making available in the state any system "specifically programmed" to deliver what would be professional mental or behavioral health care if a human did it. It also bans claims, explicit or implied, that an AI system, avatar or feature is a therapist, counselor or psychiatrist. The civil penalty is up to $15,000 per violation. Licensed providers may not use AI to deliver care directly to patients. Public schools may not use AI to perform the mental-health duties of school counselors, psychologists or social workers.

Five more states followed in 2026:

• Maine — LD 2082 (Public Law chapter 687), approved by the governor on April 13, 2026. It mirrors Illinois's "provide, advertise or otherwise offer" ban and makes violations an offense under the Maine Unfair Trade Practices Act. It was not enacted as an emergency measure, so under Maine's general rule it took effect 90 days after the Legislature adjourned its 2026 session.
• Tennessee — SB 1580 / HB 1470 (Public Chapter 647), signed April 6, 2026, effective July 1, 2026. It prohibits developing or deploying AI that advertises or represents itself as a qualified mental health professional, with a $5,000 civil penalty per violation under the state Consumer Protection Act. It passed 32-0 in the Senate and 94-0 in the House.
• Colorado — HB26-1195, "Psychotherapy Artificial Intelligence Restrictions," signed June 3, 2026, effective August 12, 2026. Licensed professionals may not use AI to conduct therapy without "synchronous, real-time interaction" between the clinician, the AI system and the client. Presenting AI output as equivalent to a licensed professional's services is treated as a deceptive trade practice.
• Vermont — H.816 (Act 156), effective June 17, 2026. A corporation or entity may not provide, advertise or offer mental health services, including through AI, unless a mental health professional provides them or they are part of an approved institutional review board (IRB) study. Violations fall under the state's Consumer Protection Act, which gives the attorney general and private parties their existing remedies.
• Rhode Island — S 2197 / H 7349, the Oversight of Artificial Intelligence Technology in Mental Health Care Act, sponsored by Sen. Lori Urso and Rep. Tina Spears. The General Assembly announced on June 23, 2026 that it had been signed into law. The final amended Senate text repeats the "provide, advertise, or otherwise offer" ban and bars AI from making independent therapeutic decisions or determining treatment plans. The text says it takes effect upon passage.

How do the state AI therapy laws compare?

The table below summarizes each law from its enacted or legislature-published text as of September 2026. "License model" means the law restricts who may deliver therapy. "Product-safety model" means the law sets rules for how a chatbot must behave.

State	Law	In force	Model	Core restriction	Enforcement
Illinois	HB 1806 (WOPR Act)	Aug 1, 2025	License	No offering therapy via AI unless a licensed professional conducts it; AI cannot detect emotions or talk to clients therapeutically	IDFPR; up to $10,000 per violation
Nevada	AB 406	Jul 1, 2025	License + product	No AI "specifically programmed" to provide mental health care; no therapist claims; school limits	Up to $15,000 per violation
Utah	HB 452	May 7, 2025	Product safety	Mental health chatbots allowed, with AI disclosure, ad and data limits	Division of Consumer Protection; up to $2,500 per violation
Maine	LD 2082	90 days after 2026 adjournment	License	No offering therapy via AI unless a licensed professional provides it	Unfair Trade Practices Act; licensing board
Tennessee	SB 1580	Jul 1, 2026	Representation	No AI that claims to be a qualified mental health professional	Consumer Protection Act; $5,000 per violation
Vermont	H.816 (Act 156)	Jun 17, 2026	License	No entity offering mental health services via AI unless a professional provides them or an IRB study	Consumer Protection Act; licensing boards
Rhode Island	S 2197 / H 7349	On passage (Jun 2026)	License	No offering therapy via AI unless licensed; AI cannot make therapeutic decisions	Executive Office of Health and Human Services
Colorado	HB26-1195	Aug 12, 2026	License	No AI therapy without real-time clinician involvement; no passing AI off as a professional	Deceptive trade practice

Which states regulate mental-health chatbots without banning them?

Utah wrote the most detailed rulebook for keeping these apps on the market. HB 452, effective May 7, 2025, defines a "mental health chatbot" as generative AI that holds therapy-like conversations and that its supplier represents, or a reasonable person would believe, can help treat mental health conditions. Scripted meditation tools are excluded. Covered chatbots must disclose they are AI before first access, after seven days of inactivity, and whenever asked. Suppliers may not sell or share users' health information or conversation inputs, or use them to target ads. A supplier that files and follows a written safety policy with the Division of Consumer Protection gains an affirmative defense against unlicensed-practice claims if the policy commits it to involve licensed therapists, test outputs, respond to acute risk in real time, and prioritize "user mental health and safety over engagement metrics or profit."

California and New York regulate "companion" chatbots generally, not therapy apps specifically. California's SB 243, approved on October 13, 2025, requires operators to disclose that a bot is artificial when a user could be misled, send minors break reminders every three hours, keep a protocol that refers users expressing suicidal ideation or self-harm to crisis services, and, starting July 1, 2027, file annual reports with the state Office of Suicide Prevention. Users harmed by violations can sue for at least $1,000 per violation. New York's General Business Law Article 47, in effect since November 5, 2025 according to Governor Kathy Hochul's letter to operators, requires crisis-detection protocols and "not a human" notices at the start of a session and every three hours.

Federal activity so far has been fact-finding rather than rulemaking. On September 11, 2025, the Federal Trade Commission issued 6(b) orders to seven companies (Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and xAI) asking how they test, monitor and monetize companion chatbots and how they limit children's access.

What actually triggers a ban: the technology or the claim?

Mostly the claim. Read side by side, the ban statutes share one design choice: they hang liability on offering, advertising or representing a service as therapy, not on whether a language model is involved. Illinois and Rhode Island exempt "self-help materials and educational resources" that "do not purport to offer therapy or psychotherapy services." Maine's ban has no such carve-out; its only exception is IRB-approved research. Nevada exempts self-help products that do not claim to provide professional care. Tennessee's law is almost entirely a representation rule. Colorado carves out non-diagnostic self-help, coaching and meditation tools, as long as they clearly say they do not replace clinical care.

Our analysis: this creates an odd incentive. The most exposed apps are specialist products marketed as clinically grounded AI therapy. General-purpose and companion chatbots that avoid the word "therapy" mostly face only the lighter disclosure and crisis-protocol rules in California and New York. Lawmakers drew the line around the label, not the use. Future bills could borrow Utah's "a reasonable person would believe" test, which reaches products based on how users perceive them.

What can licensed therapists still do with AI?

More than the "ban" framing suggests. Every license-model state keeps a lane open for clinicians:

• Administrative work is broadly allowed: scheduling, reminders, billing and insurance claims in Illinois, Maine and Rhode Island; records and operational data analysis in Nevada.
• AI note-taking is allowed with written consent. Illinois and Maine let clinicians use AI for supplementary support such as therapy notes when a session is recorded or transcribed, but only after written notice and consent. Rhode Island applies a similar written-consent rule to companion-style AI tools used in recorded or transcribed sessions. All three state that accepting a general terms-of-use agreement does not count as consent. Maine adds that clients must be told how session data will be stored, retained, used for training and deleted, and that a clinician cannot refuse treatment because a client declines AI. This is the same category of tool covered in our look at ambient clinical AI scribes like DAX, Abridge and Suki.
• Supervised use varies by state. Colorado permits AI in therapy only with synchronous, real-time clinician involvement. Rhode Island's final amended text ties any direct therapeutic communication by AI to an established clinician-patient relationship the patient has consented to, with the clinician keeping responsibility for clinical judgment and oversight. Illinois and Maine bar AI from therapeutic communication with clients outright.
• FDA-cleared products get an exit ramp. Vermont allows software medical devices and digital therapeutics that are authorized, cleared or approved by the US Food and Drug Administration when a mental health professional prescribes or recommends them. Rhode Island's text exempts AI tools cleared by the FDA or another federal health-AI regulator.

What should mental-health app makers do now?

This is general information, not legal advice. The statutes differ in scope and definitions, so companies should have counsel review how any one of them applies to their product.

1. Audit your language before your model. "Therapy," "therapist" and "counselor" in marketing, onboarding or an avatar's persona create the most exposure. Nevada's law explicitly covers "any component, feature, avatar or embodiment."
2. Decide whether to geofence. In license-model states, a product positioned as AI therapy likely needs a licensed clinician in the loop; the alternative is not offering it there. Utah defines its users by where they are when they use the app, a hint that location at the time of use is the likely test.
3. Treat the clinician-in-the-loop as architecture. Colorado's real-time-involvement rule and Rhode Island's established-relationship rule reward models where AI supports a licensed provider, as employer platforms in our comparison of Lyra, Spring and Modern Health already do.
4. Price in the FDA route. Vermont and Rhode Island give FDA-regulated software a clearer path than consumer apps.
5. Build to Utah's checklist anyway. Its affirmative-defense list (clinician involvement, pre-launch testing, harm reporting, real-time crisis response) is one of the most detailed public statements of what a regulator expects.

What do these laws mean for people using AI for mental health?

If you live in a ban state, talking to a chatbot is not illegal. The obligations fall on companies and clinicians. Some apps may block your state or drop therapy language. If you see a therapist, several states now require written notice before AI transcribes your sessions, and let you say no.

The laws also make a point that is easy to lose in the product debate: these tools are not licensed clinicians, and no state has said they are. Utah's law says outright that nothing in it recognizes a chatbot as a licensed mental health therapist. For context on how individual products have positioned themselves, see our earlier review of AI mental health chatbots including Wysa, Woebot and Replika. Anyone in crisis should contact local emergency services or a licensed professional rather than rely on an app.

What to watch next

First, enforcement: the first IDFPR hearing or attorney general action against a named AI company will set the real boundary of "offer therapy." Second, under Act 156, Vermont's Artificial Intelligence Advisory Council must report to lawmakers by January 15, 2027, on regulating mental health professionals' use of AI. Third, the template: the 2026 class mostly copied Illinois while adding clinician-involvement conditions and FDA carve-outs, and that hybrid looks like the likeliest model for the next wave.

The bottom line: regulators in these states ask not "Is this AI?" but "Is this being offered as therapy?" Builders who answer clearly, through marketing, clinical supervision or FDA clearance, can still operate.

Frequently Asked Questions

Which states have banned AI therapy chatbots?

As of September 2026, Illinois (HB 1806) and Nevada (AB 406) have laws in force from 2025. Maine (LD 2082), Tennessee (SB 1580), Vermont (H.816), Rhode Island (S 2197 / H 7349) and Colorado (HB26-1195) followed in 2026. Each restricts offering AI as therapy unless a licensed professional provides or supervises the service, or bars AI from claiming to be a mental health professional. The details and penalties differ by state.

Is it illegal to use ChatGPT or another chatbot for emotional support in these states?

No. The laws regulate companies that provide or advertise therapy services, and licensed clinicians who use AI. They do not penalize individuals for what they type into a chatbot. General-purpose assistants that do not market themselves as therapy mostly fall outside the ban language, although companion-chatbot laws in states like California and New York still require crisis protocols and AI disclosures.

Can my therapist use AI to take session notes?

In most of these states, yes, with conditions. Illinois and Maine allow AI for supplementary tasks like therapy notes, but when sessions are recorded or transcribed, the clinician must tell you in writing that AI will be used and for what purpose, and get your consent. Rhode Island has a similar consent rule for companion-style AI tools. A general terms-of-service click does not count. In Maine, a clinician cannot refuse to treat you if you decline.

How is Utah's law different from Illinois's?

Illinois prohibits offering therapy through AI unless a licensed professional conducts it. Utah allows mental health chatbots to operate but requires AI disclosures, bans selling or sharing users' health data and conversation inputs, restricts in-chat advertising, and gives suppliers a defense against unlicensed-practice claims if they file and follow a safety policy with the state's Division of Consumer Protection. One restricts who may provide therapy. The other regulates how the product behaves.