AI Therapy Laws by State: The Burden Lands on Your License

If you have gone looking for AI therapy laws by state lately, you already know the count keeps moving. In 2026 alone, 43 states introduced more than 240 health-AI bills, and 36 states filed 70-plus chatbot bills in the first quarter, according to the Manatt Health AI Policy Tracker (April 9, 2026). Across the wider 2024-2026 window, only Wyoming and North Dakota have sat the whole thing out. Somewhere in that pile is a rule that now governs how you write your notes.

Here’s the part almost nobody says out loud. Most of these laws regulate the companies selling AI tools, not the people holding licenses. That sounds like good news for a practice. It isn’t. A vendor can be fully compliant in every state it ships to, and the clinician using that product still owns every output it produces. That includes most of what sits in a typical private practice AI stack: the scribe, the intake summarizer, the scheduling assistant.

Quick answer: most state AI-in-mental-health laws regulate vendors, but vendor compliance creates no safe harbor for the licensee. Maine’s new law puts full responsibility for all AI interactions, outputs and data use on the licensed professional. Vendor-facing laws are enforced by attorneys general, and the penalty is money. Practice-facing laws are enforced by licensing boards, and the penalty is your license.

Key Takeaways

• Of the major state AI-therapy laws, only Illinois, Maine, Texas and Arizona attach duties to the licensee. The rest regulate vendors.
• Maine’s LD 2082 takes effect July 29, 2026, and contains no dollar penalty. The widely repeated $10,000 figure belongs to Illinois HB 1806.
• Enforcement channel matters more than fine size: a board action follows you, a corporate fine follows a balance sheet.
• 43 states are legislating. Six Counseling Compact states actually issue privileges (Counseling Compact, June 2026).

What do AI therapy laws by state actually regulate?

Most of them regulate software companies. Of the states covered here, four impose duties on licensed clinicians and the rest land on developers, deployers, chatbot suppliers and “AI providers.” MultiState counted 12 states with enacted companion-chatbot laws as of June 26, 2026, and nearly all of those target the operator, not the practice.

That distinction gets flattened in most coverage, which is how a lot of practice owners ended up worried about the wrong things. California’s SB 243 and New York’s General Business Law Article 47 both require suicide and self-harm protocols with crisis referrals. Those duties sit on companion-chatbot operators. Neither statute creates an obligation for a therapy practice. They belong to the same fight as the litigation over how general-purpose chatbots handle users in crisis. That’s a product-design question, not a clinical one.

Nevada’s AB 406 bars AI providers from offering or programming AI that constitutes the practice of mental or behavioral healthcare. Utah’s HB 452 makes mental health chatbot suppliers disclose AI status before access, after seven days of non-use, and on request. Both are vendor statutes. Neither one reaches a licensee.

So why does any of this land on a small practice? Because four states went the other direction, and because the ones that did chose the enforcement channel with the sharpest teeth.

AI therapy laws by state: what each one actually does

Effective dates run from July 2025 through July 2027, which makes this a rolling compliance calendar rather than a single deadline. The table below sorts every enacted provision by who it actually binds. Four of these attach to the licensee. That column is the one to read first.

State	Law	What it does	Effective	Binds
Illinois	HB 1806 (WOPRA)	Bars AI from independent therapeutic decisions, direct therapeutic communication, or generating treatment plans without licensed review. Administrative and supplementary support permitted. Up to $10,000 per violation, IDFPR-enforced.	Aug 4, 2025	Both
Nevada	AB 406	Bars AI providers from offering AI that constitutes mental or behavioral healthcare, and from claiming AI can provide it. Up to $15,000 per violation.	July 1, 2025	Vendor
Utah	HB 452	Mental health chatbot suppliers must disclose AI status before access, after 7 days of non-use, and on request. Ad and data-sale limits. Up to $2,500 per violation.	May 7, 2025	Vendor
California	AB 489 (Ch. 615)	Bars AI systems and their developers or deployers from using titles, post-nominals or icons implying licensed care. Enforced by healing-arts licensing boards.	Jan 1, 2026	Vendor
California	SB 243	Companion-chatbot operators must maintain and publish a protocol preventing suicide and self-harm content, with crisis referral. Annual reporting begins July 1, 2027.	Jan 1, 2026	Vendor
New York	GBL Art. 47	Operators must disclose AI status at start and every 3 hours, detect suicidal ideation, and refer to crisis providers. Up to $15,000 per day, AG-enforced.	Nov 5, 2025	Vendor
Maine	LD 2082, PL Ch. 687	Non-licensees may not provide or advertise therapy, including via internet-based AI (a UTPA violation). Licensees may use AI for administrative and consented supplementary support only, with recording or transcription plus written disclosure. Full responsibility stays with the clinician.	July 29, 2026	Both
Texas	SB 1188	Practitioners may use AI diagnostically only within license scope, with all AI-generated records reviewed and use disclosed to patients.	Sept 1, 2025	Licensee
Texas	HB 149 (TRAIGA)	Providers must inform patients or guardians of AI use in treatment by date of service. 60-day cure period after AG notice.	Jan 1, 2026	Licensee
Arizona	Board of Behavioral Health Examiners rule	Behavioral health professionals must obtain and document informed consent before services involving AI, machine learning, or human-simulation modality.	Jan 1, 2027	Licensee
Tennessee	SB 1580	Prohibits marketing or advertising AI as a qualified mental health professional.	July 1, 2026	Vendor
Oregon	SB 1546	Operators must implement evidence-based protocols to detect and respond to suicidal ideation.	Jan 1, 2027	Vendor
Idaho / Nebraska	Conversational AI Safety Acts	AI disclosure plus crisis protocols. Nebraska also bars representing AI as licensed healthcare.	July 1, 2027	Vendor
Delaware	HB 191	Bars nonhuman entities from licensure as healthcare professionals.	Effective date not confirmed at publication	Vendor

Read the Binds column top to bottom and the pattern is obvious. Vendor-facing rows outnumber practice-facing ones by roughly two to one. If bill count were the story, practices would be the ones getting off easy. Bill count is not the story.

One row is worth a closer look because it shows how narrow most of these statutes really are. Tennessee’s SB 1580 prohibits marketing or advertising AI as a qualified mental health professional. That’s a truth-in-advertising rule aimed at products. It says nothing about whether a licensed therapist in Memphis can run a scribe.

If my AI vendor is compliant, is my practice covered?

No. Vendor compliance and licensee compliance run on separate tracks, and Maine says so in the statute. Under 32 MRSA §2113, a licensee using AI must maintain “full responsibility for all interactions, outputs and data use associated with the use of artificial intelligence.” No provision in that law transfers any part of that responsibility to the software company.

Think about what that means in practice. A scribe vendor can satisfy Nevada, Utah, California and New York in full. Its compliance page can be immaculate. The Maine clinician running that product still owns every summary it drafts, every inference it makes, and every disposition of the session data behind it. That last item is the one most practices have never actually traced. Who owns therapy data on an AI platform is now a question the statute expects you to answer in writing.

There is a second layer here, and it’s the one that actually decides who pays. Amy Worley of BRG told TechTarget (January 27, 2026) that the vendor-versus-provider liability split is set by contract, not by statute. Nothing in these laws allocates risk between the two of you. Your terms of service do.

Now picture a solo practice negotiating that contract. There is no negotiation. It’s a click-through, drafted by the party with the legal department, and the default allocation is the one you’d expect: vendor writes the terms, practice carries the risk. That is the offloading, and it’s documented rather than alleged.

Worley also named the underlying mess plainly: it is “so convoluted. There are multiple definitions of even what AI is in the existing state laws.” A practice can’t write one policy when the word that policy governs means different things in different jurisdictions.

A board action costs more than a corporate fine

The enforcement channel, not the dollar figure, determines what a violation actually costs you. Vendor-facing laws route through attorneys general and consumer-protection divisions: up to $15,000 per violation in Nevada, $2,500 in Utah, $15,000 per day in New York. Practice-facing laws route through licensing boards, where the monetary ceiling under Maine’s board-discipline statute (10 MRSA §8003) is a fraction of those figures.

On paper, that looks like clinicians got the lighter end. Read it again. A corporate fine is a line item. It gets paid, booked, and forgotten by the next quarter. Nobody asks a software company about it during credentialing.

A board action is a different species of consequence. It’s reportable and durable. It surfaces on license verification, payer applications and privileging paperwork for years, and in most states it’s public. The dollar amount is an order of magnitude smaller. The career cost isn’t close. Cheaper on paper, more expensive in life.

This asymmetry is why the “most of these laws target vendors” framing is so misleading. The laws that reach practices are fewer, quieter, and enforced by the one body that can end a career. Illinois’ $10,000 penalty comes with IDFPR jurisdiction over the license itself. Texas requires review of all AI-generated records. Arizona requires documented informed consent before any AI-involved service starting January 1, 2027. None of those are money problems.

The $10,000 Maine error is already circulating

One correction worth making loudly, because the mistake is spreading through aggregator coverage. Maine’s law carries no monetary penalty. I read the chaptered text of Public Law Chapter 687 in full. There is no dollar figure anywhere in it.

The $10,000 belongs to Illinois HB 1806, per the IDFPR press release of August 4, 2025. Maine routes unlicensed AI therapy through the Unfair Trade Practices Act, where a $10,000 figure attaches to violating an injunction, not to the underlying act. Licensee violations go to the board.

The effective date is getting mangled too. Aggregator sites list April 13, 2026. That’s the governor’s approval date. The operative date is July 29, 2026, derived from Maine’s constitutional 90-day rule running from sine die adjournment on April 29, 2026. Days from now, not months ago.

Which state’s rules apply when you practice across state lines?

The state where your client is sitting. Established telehealth licensure practice holds that a clinician must generally be licensed where the patient is physically located at the time of service, and these AI provisions live inside professional-practice titles. Maine’s sit in Title 32, the licensing title. They follow the same attachment point.

That single principle is why fragmentation lands on the practice instead of the vendor. A software company ships one product nationally and complies with the strictest applicable rule. A practice with clients in six states inherits six regimes, six consent standards, and six sets of definitions for the same word. It’s the same attachment point that governs everything else about setup. Opening a practice in New York never really means opening it only in New York.

And this is not a niche problem anymore. In the APA Practitioner Survey reported by the American Telemedicine Association, 88% of responding psychologists were practicing in hybrid or fully remote arrangements. That data is from 2023 and it’s aging. The same survey named cross-jurisdictional issues as a top barrier to sustaining telehealth. That was before any of these AI laws existed.

Multi-state telehealth stopped being an edge case around 2021. The regulatory architecture never caught up to that fact, and now it’s issuing AI rules on top of a foundation that was already cracked.

Do the licensure compacts fix this?

Not yet, and the gap is the sharpest number in this whole story. The Counseling Compact reports six states actually issuing privileges as of June 2026: Arizona, Georgia, Indiana, Louisiana, Minnesota and Ohio. Another 32 states plus DC are still “actively completing the steps.” That’s the official .gov figure, not an estimate.

Compacts were the answer to exactly this problem. One privilege, many states, less administrative drag on clinicians who see clients across lines. The Social Work Licensure Compact is reported at roughly 35 states and is still not operational for licensees.

Set the two timelines beside each other. Forty-three states moved on health AI in a single 2026 session. Six states can issue a counseling privilege after years of work. Regulation scaled at legislative speed. Relief scaled at bureaucratic speed. The clinician absorbs the difference, and no mechanism anywhere compensates them for it. That pattern should look familiar to anyone who has watched reimbursement quietly reprice their work without a single conversation about it.

That’s the system critique, and it doesn’t require anyone to be a villain. It only requires that the fast-moving part of government and the slow-moving part never talk to each other, which is the normal condition of American health regulation.

Isn’t this just regulation based on fear?

It’s a fair challenge and worth stating honestly. The Reason Foundation has argued that Nevada’s ban on AI therapists is “regulation based on fear rather than analysis,” and that blanket prohibitions foreclose tools that might expand access in a country with a serious clinician shortage. Reasonable people hold that position, and the access problem it points at is real.

Here’s the thing, though. The deregulatory critique and the fragmentation critique are pointing at the same failure from opposite sides. Nobody built a coherent national floor for AI in behavioral health. So the country got a dozen incoherent ones instead, with 240-odd bills queued behind them. Whether you think the rules are too strict or too scattered, the smallest actor in the system absorbs the cost of the incoherence. That’s the part both camps should be able to agree on.

Your compliance checklist before the deadlines hit

Start with an inventory, because AI therapy laws by state attach to where your clients sit, not where your office is. Which AI-touching tools are in the workflow, what do they do with session content, and in which states are your clients located when the tool runs? Arizona’s consent-documentation rule arrives January 1, 2027, so the runway is short but real.

A few things follow from the statutes themselves rather than from anyone’s opinion. Maine’s supplementary-use path requires the session to be recorded or transcribed, plus written disclosure covering what the AI is, its purpose, and how session data is stored, retained, used for training, and deleted. Consent has to be explicit. Waivers are void and unenforceable.

Two provisions deserve special attention because they change practice design, not just paperwork. Maine bars denying treatment to a client who refuses AI consent, which means your workflow needs a working no-AI path, not just a consent form. And Texas SB 1188 requires review of all AI-generated records, which is an operational commitment rather than a checkbox.

Then read your vendor contract with the liability question in mind. You probably can’t change the terms. You can at least know what you signed and price the risk accordingly.

None of this is legal advice, and I’m a clinician rather than a lawyer. Rules change, definitions differ by state, and your board may read its own statute differently than a law-firm alert does. Confirm anything here with your licensing board and your own counsel before you change how you practice. If it helps to talk through what this means for the shape of your practice, you can book a call.

FAQ

Which states currently restrict AI in therapy? As of July 2026, Illinois and Nevada bar AI from delivering mental health treatment; Maine restricts licensees to administrative and consented supplementary use starting July 29, 2026; Utah, California, New York, Oregon, Idaho, Nebraska and Tennessee regulate the AI products themselves; and Texas and Arizona impose disclosure and consent duties directly on providers. Twelve states have enacted companion-chatbot laws as of June 2026, and 43 states introduced health-AI bills in 2026.

What does Maine’s LD 2082 actually require, and when? Effective July 29, 2026. Licensees may use AI for administrative support and, with written disclosure and explicit client consent, for supplementary support such as note preparation, but only when the session is recorded or transcribed. AI may not make independent therapeutic decisions, interact therapeutically with clients, or produce treatment plans without licensee review. The clinician retains full responsibility for all AI outputs. A client cannot be denied treatment for declining AI consent, and any waiver of these protections is void.

Does Maine’s law carry a $10,000 fine? No. The chaptered law contains no dollar figure. That $10,000 figure belongs to Illinois’ HB 1806. In Maine, offering AI therapy without a license is a Maine Unfair Trade Practices Act violation, while a licensee’s misuse of AI is referred to their professional board for disciplinary action.

If my AI vendor says it is compliant, is my practice covered? No. Most of these statutes regulate vendors and licensees on separate tracks. Maine’s law expressly places full responsibility for all AI interactions, outputs and data use on the licensed professional. Vendor compliance and licensee compliance are distinct obligations, and the split in liability is set by your contract, not by statute.

Which state’s rules apply if I see clients across state lines? Under established telehealth licensure practice, a clinician is generally governed by the rules of the state where the client is physically located during the session. Because these AI provisions sit inside professional-licensing statutes, a practice serving clients in several states is generally answering to several sets of rules simultaneously.

Do the licensure compacts solve this? Not yet. The Counseling Compact has six states actually issuing privileges, with 32 states plus DC still working toward it. The Social Work Licensure Compact is reported at roughly 35 states but is not yet operational for licensees. The regulation is arriving faster than the portability infrastructure meant to make it manageable.

Sources

Maine Legislature, LD 2082 / HP 1397 status and chaptered text (Public Law Ch. 687) and the chaptered law PDF, read in full: no monetary penalty, licensee responsibility language, consent and disclosure requirements. Maine Title 10 §8003, board disciplinary authority and civil penalty ceiling. Maine Title 5 Ch. 10, Unfair Trade Practices. Illinois Department of Financial and Professional Regulation, press release on HB 1806 (August 4, 2025), $10,000 per violation. California AB 489 (Ch. 615, 2025) and the SB 243 signing announcement. Counseling Compact (June 2026), six privilege-issuing states, 32 states plus DC in progress. Social Work Licensure Compact. Manatt Health AI Policy Tracker (April 9, 2026), 43 states and 240-plus health-AI bills, 36 states and 70-plus chatbot bills in Q1. MultiState, Vol. 105 (June 26, 2026), 12 states with enacted companion-chatbot laws. American Telemedicine Association on the APA Practitioner Survey, 88% hybrid or remote practice (2023 data). Holland & Knight, states continuing to regulate AI in healthcare (May 26, 2026), Texas, Arizona, Tennessee, Oregon, Idaho, Nebraska and Delaware provisions. Akin Gump on Texas HB 149 and SB 1188. Wilson Sonsini on Nevada AB 406 and Utah HB 452. Fenwick on New York’s AI companion safeguard law. TechTarget, how providers can navigate the patchwork of state health AI laws (January 27, 2026), Amy Worley of BRG on contract-set liability and definitional inconsistency. Counter-perspective: Reason Foundation on Nevada’s AI therapist ban. Figures current as of July 24, 2026.

Disclaimer

This article is for educational and informational purposes only. It does not constitute legal, medical, clinical, or therapeutic advice, and reading it does not create an attorney-client or therapist-client relationship with Matthew Sexton, LCSW or Mental Wealth Solutions, Inc.. Although the author is a licensed clinical social worker, nothing here is a legal opinion on any statute, rule, or contract.

State laws, board rules, effective dates, and regulatory interpretations vary by jurisdiction and change frequently, and provisions described here may be amended, delayed, or interpreted differently after this article is published. Statutory summaries are condensed and are not a substitute for reading the operative text. Before changing how your practice uses any AI tool, confirm the requirements with your own licensing board, your malpractice carrier, and qualified legal counsel licensed in the states where your clients are located.

If you are in immediate emotional crisis, you can reach the 988 Suicide & Crisis Lifeline by calling or texting 988 (US). If you are experiencing domestic violence or are in physical danger, contact the National Domestic Violence Hotline at 1-800-799-7233 or visit thehotline.org. In a life-threatening emergency, call 911.