AI therapy laws by state

State AI therapy laws regulate different activities. Some restrict a clinician’s use of AI in treatment. Others govern companies offering chatbots, prohibit misleading professional claims, or require safeguards when a chatbot user expresses distress. A product can fall within more than one category. A disclosure that a tool uses AI does not, by itself, make its therapeutic use lawful.

Version: 2026-09. Last reviewed: September 12, 2026. Review cadence: monthly. Next scheduled editorial review: October 12, 2026. This is a dated editorial reference, not a continuously updated legal database. The review date describes this edition; it does not promise that subsequent reviews have already occurred.

What does this state AI-law register cover?

This register covers laws and proposals specifically addressing AI in mental health services, mental health chatbot products, and conversational or companion chatbots whose rules address mental health, professional representation, or self-harm. It also identifies directly relevant restrictions on payment for AI-delivered psychotherapy and healthcare AI disclosure provisions that reach licensed mental health services. General privacy, recording, licensure, consumer protection, insurance, and health information laws remain separate layers of analysis.

The tables distinguish an enacted law from a pending bill and an effective provision from a future obligation. Sources are official statutes, chaptered acts, bill texts, and legislative histories unless a verification limitation is expressly identified. Proposed dates in unenacted bills are not compliance deadlines.

Coverage limitation: a complete negative finding for every unlisted bill in all 50 state legislatures could not be established in this review. The coverage ledger below identifies unresolved jurisdictions and retrieval gaps. Do not read an omitted bill, an inaccessible source, or an unsuccessful search as proof that a state has no relevant law. Secondary trackers were used for discovery, not as substitutes for checking a measure’s session, final text, and disposition.

Which enacted AI therapy and chatbot laws should clinicians check?

The practical duty depends on whether you are treating a patient, operating a chatbot, developing a product, or performing more than one role. “Applies to” describes the relevant statutory actor, not everyone who mentions AI in a session. Each row links to the controlling text or identifies the remaining text-verification gap.

The California January 1 commencement dates follow the ordinary rule for nonurgency statutes enacted in the preceding regular session; the separately stated operative dates govern delayed provisions. Confirm the current chaptered text when implementing an amendment. Signing SB 1119 did not make all its July 1, 2027 duties immediately operative.

What must Illinois clinicians do before using AI?

Illinois distinguishes support tasks from therapeutic activity. Its Act prohibits independent therapeutic decisions, direct AI therapeutic communication with clients, and emotion or mental-state detection. Recommendations and plans require professional review and approval. Permitted administrative or supplementary use leaves responsibility with the professional.

When AI records or transcribes a session for supplementary support, provide written information identifying AI use and its specific purpose, then obtain the statutory affirmative, written, revocable consent. A general terms-of-use acceptance is insufficient. The Act defines the licensed professions it covers; do not silently extend that definition to every healthcare license. These provisions and the August 1, 2025 effective date appear in Public Act 104-0054.

Practical duty: classify each feature separately. A tool that drafts a note and also sends personalized therapeutic messages presents two different legal questions. Review the note yourself and disable functions the Act prohibits. Patient consent does not override a prohibited function.

What must Nevada clinicians check about administrative AI?

Nevada’s provider restriction is broader than a ban on a chatbot pretending to be a therapist. NRS 629.610 restricts using AI to provide professional mental and behavioral healthcare, subject to listed administrative exceptions and the specified school-policy exception. It requires independent accuracy review for billing-related outputs and session file or note functions identified in subsection 2(c) and 2(e). See the operative provider statute.

The supplier rule in NRS 433.567 separately concerns systems specifically programmed to provide experiences that would constitute professional care if delivered by a natural person, and covered representations. Avoid replacing those definitions with “all chatbots are banned.” The legislative effective-date record establishes July 1, 2025 for the substantive AB 406 provisions.

Practical duty: map your proposed use to an actual exception. An AI-generated diagnostic interpretation is not made administrative merely by placing it inside a progress-note template.

What must Texas clinicians disclose about healthcare AI?

Texas HB 149 includes a healthcare disclosure provision, even though the broader law is not limited to therapy. Section 552.051(f) requires the service provider to disclose AI use to the recipient or personal representative by the date service or treatment first occurs. In an emergency, disclosure follows as soon as reasonably possible. The notice must be clear, conspicuous and in plain language. Section 552.052 separately prohibits developing or deploying AI with the intentional aim of inciting self-harm, harm to others, or criminal activity. See the enrolled Act.

The official legislative history confirms signature on June 22, 2025 and effectiveness on January 1, 2026.

Practical duty: include the healthcare disclosure analysis in your tool review. Do not assume that a provision located in a general AI statute applies only to technology companies. Providing notice does not establish competence, authorize unlicensed treatment, or resolve confidentiality duties.

What do Utah’s mental health chatbot rules require?

Utah regulates defined mental health chatbot suppliers through restrictions on sharing health information and user inputs, advertising practices, and AI identity disclosures. Disclosure is required before access, after a gap exceeding the statutory seven-day interval, and when the user asks whether AI is involved. Limited sharing exceptions do not create a general permission to sell sensitive conversations. The current Chapter 72a text includes the 2026 amendments and distinguishes generative therapy-like tools from scripted exercises or referral-only systems.

A separate supplier policy process relates to an affirmative defense under §58-60-118. Filing a policy is not a professional license or proof of clinical effectiveness. The Utah licensing division’s registration page links that process to the governing statutes.

Practical duty: ask what user data leaves the tool, who receives it, and why. Obtain the actual policy and contractual terms. Do not substitute the vendor’s registration claim for an assessment of the intended clinical use.

Which California requirements are current, future, or still proposed?

SB 243 governs defined companion chatbot operators. Its current framework includes an identity notice where a reasonable person could mistake the chatbot for a human, a published self-harm protocol, and additional protections for known minors. Reporting beginning July 1, 2027 is a separate date from the law’s initial commencement. See the chaptered SB 243 text.

SB 1119 was signed on September 10, 2026. It restructures child safeguards and adds risk assessment, data-use and audit provisions with distinct implementation dates. Its principal child safety duties start July 1, 2027; the initial audit deadline is generally January 1, 2029, with revenue-dependent provisions. It also addresses attempts to diagnose or treat child users, with a specified FDA and HIPAA exception. Use the final Adam’s Law text, not an earlier committee draft.

SB 867 addresses toys containing companion chatbots. A clinic considering an interactive toy should examine the statutory product definition and sales restrictions, rather than treating the device as exempt because it sits in a playroom. See SB 867.

Practical duty: separate operating a companion product from using an administrative tool. SB 903, the therapy-specific proposal, remains pending as described below. Its proposed duties must not be presented to patients or staff as already enacted California law.

What does Colorado require when AI enters psychotherapy?

Colorado’s HB 26-1195 is effective August 12, 2026. It requires synchronous, real-time interaction involving the professional, AI, and client for covered AI therapeutic communication, and review and approval of recommendations or plans. It addresses initial-contact disclosures and advance written informed consent for AI recording or transcription. Its exclusions and exceptions matter, including specified educational, research, support and non-diagnostic uses. The enacted summary and signed text should be read together.

HB 26-1263 separately regulates conversational AI operators, including professional representations, with major duties commencing January 1, 2027. HB 26-1139 includes restrictions on payer reimbursement for psychotherapy conducted directly by AI and human review in utilization review. See the operator act and the healthcare act.

Practical duty: review the between-session workflow as well as the session itself. A clinician reviewing a chatbot conversation the next morning is not the same arrangement as synchronous professional participation. Check payment rules before billing any service involving an AI component.

What does Maine require beyond a general AI disclosure?

Maine’s LD 2082 adds parallel provisions across specified professional chapters, including social work, psychology, counseling, medicine and nursing. It restricts autonomous therapeutic decisions and direct therapeutic communication, requires review of recommendations and plans, and regulates support uses. The written disclosures for supplementary support address purpose, storage, retention, training use and deletion of session data. The text uses a specific recording or transcription condition that should be reviewed carefully before assuming every support workflow qualifies.

A client cannot be denied therapy solely for declining the covered supplementary AI use. Read Public Law Chapter 687. The legislature’s session notice establishes July 29, 2026 as the general effective date for these nonemergency laws.

Practical duty: offer a workable consent-refusal process. “The platform requires it” does not answer the patient’s statutory choice. Separately check LD 1727 when a practice uses an AI-facing consumer interface; its transparency rule is not permission to provide automated psychotherapy.

How does Rhode Island distinguish clinical use from companion products?

The therapy act preserves human professional responsibility while distinguishing independent decisions, treatment planning, and therapeutic communication within an established professional relationship. It includes a specific written-consent condition for described AI uses when a session is recorded or transcribed. It also addresses a client’s independently chosen AI use: the professional may discuss it and must attend to confidentiality, safety, and risks. See H 7349A’s operative text.

The separate companion law has identity and harm-response duties and takes effect January 1, 2027. Its enactment is confirmed by the S 2195 history; the final text controls its scope.

Practical duty: do not import Illinois’s exact rule into Rhode Island. Read the established-relationship language, consent condition, and exceptions. Confirm the therapy act’s exact chapter and signing date before using that date in a compliance notice.

What is Tennessee’s actual restriction on AI therapy claims?

Tennessee’s Public Chapter 647 prohibits a developer or deployer from advertising or representing that an AI system is, or can act as, a qualified mental health professional. The official HB 1470 history confirms enactment and July 1, 2026 effectiveness. The one-page bill text supplies the prohibition. The signed chapter PDF was blocked during this review; confirm the final chapter against that text before relying on a precise enforcement interpretation.

Two other measures are easily misreported: SB 1493 and HB 1946’s companion became study requirements, not the original proposed chatbot safety regimes.

Practical duty: review advertising, onboarding, product names and chatbot statements. Avoid describing the enacted representation restriction as a blanket prohibition on every administrative AI tool.

What is verified about Vermont’s mental health AI law?

Vermont enacted H 816 as Act 156. The official record confirms approval on June 17, 2026, and the indexed enacted text states effectiveness on passage. See the bill history and the enacted act.

The complete enacted PDF repeatedly failed to load. Its reported restrictions on independent AI mental health services, professional review, and research or FDA-related exceptions could not all be checked against the final text. Earlier legislative versions are not sufficient to close this gap.

Practical duty: obtain the final §7115 text before implementing an AI therapeutic workflow. Retain human responsibility for diagnosis and treatment while that review occurs. This entry confirms enactment without converting an inaccessible exception into permission to practice.

What should Connecticut clinicians check before January 1, 2027?

Connecticut’s companion rules require crisis detection and referral protocols, identity disclosures, and protections for minors. The minor mental health service provision contains specific conditions concerning design, clinical practices, accountability, accessible information and nonprofessional disclosure. It is not an unrestricted permission for a chatbot to deliver pediatric therapy. The definition also excludes certain narrowly defined healthcare support systems. See Public Act 26-15, §§4–6.

Practical duty: determine whether a proposed tool qualifies as a companion or satisfies an exclusion. Ask the operator how it distinguishes ordinary conversation from an expression requiring crisis response. General assurances about “safety filters” are not a substitute for the statutory protocol.

What does Georgia require of a clinical chatbot operator?

Georgia’s SB 540 includes identity notices at the start of interactions and specified repeat intervals, with more frequent disclosures for minors. It requires crisis protocols and limits harmful relationship features for minors. The professional-representation clause includes a qualification for operators lawfully authorized to provide the services. That qualification should be read alongside professional scope and licensure law. The signed text sets July 1, 2027 as the effective date.

Practical duty: determine whether the practice is an operator, rather than assuming all obligations sit with the underlying model developer. A clinic deploying a product under its own name should review both its clinical responsibilities and the operator definition.

What does Hawaii’s already-effective companion law require?

Hawaii’s Act 248 took effect on July 14, 2026. It addresses AI identity, self-harm response, professional mental or behavioral healthcare representations, and safeguards for minors. Its enacted definition and exclusions determine which products are covered. The governor’s approval and final text are available together in the signed act.

Practical duty: review any companion product already offered to Hawaii users. This is not a future July 1, 2027 obligation borrowed from another state’s law. Ask for the product’s current disclosures and crisis protocol, and inspect what the system actually tells users about its professional role.

What should Idaho clinicians verify before relying on S 1297?

S 1297 is reported as enacted with July 1, 2027 effectiveness, but both the official engrossed PDF and relevant code pages failed to load. The reported identity, crisis, professional-claim and minor-safety provisions therefore remain a final-text verification item in this edition. The official text endpoint is retained for the next review.

Practical duty: do not treat this entry as a determination that a specific product is permitted or exempt. Obtain the final operator definition and exceptions before adopting a vendor’s Idaho compliance representation.

What will Iowa require from conversational AI services?

Iowa’s SF 2417 applies beginning July 1, 2027. It requires disclosures, a self-harm response protocol, and protections for minors. Its professional representation provision addresses knowingly and intentionally programming or causing statements that lead a reasonable person to believe a service is designed to provide specified professional psychology or behavioral health services. Read the conditions and exclusions in the enrolled text.

Practical duty: review the product interface and generated statements, not only a disclaimer at account creation. Check whether the operator’s defined role and a product exclusion fit the actual service.

What will Nebraska’s chatbot provisions change?

Nebraska placed its Conversational Artificial Intelligence Safety Act within LB 525, alongside unrelated agricultural data provisions. The chatbot sections become operative July 1, 2027. They address identity notices, minor protections, self-harm protocols, and representations that the product provides professional mental or behavioral healthcare. The slip law, §§12–19 is the relevant source.

Practical duty: use the chatbot section’s date and definitions. A summary of the agricultural portions of LB 525 cannot establish the requirements for a clinical product, and the act’s other commencement dates should not be substituted for the chatbot date.

What does New Hampshire’s law mean for clinical AI use?

New Hampshire’s HB 143 is a child-protection measure addressing specified conduct through responsive generative communication. It is not a comprehensive therapist licensure statute. RSA 507:8-k establishes civil enforcement and liability under stated knowledge and conduct conditions; its source note gives January 1, 2026 effectiveness. Related prohibited conduct is addressed in RSA 649-B:3.

Practical duty: if a practice operates an interactive tool for children, examine its potential conduct and escalation procedures. Do not present compliance with this narrow law as approval of the tool’s diagnosis, treatment, privacy, or evidence base.

Which New York AI companion duties are already in force?

New York’s Article 47 requires a reasonable-efforts protocol for detecting and addressing suicidal ideation or self-harm, including crisis referrals. It also requires clear nonhuman identity notifications at the beginning of interaction, no more than once daily being necessary for that initial notice, and at least every three hours of continuing interaction. See §1701 and §1702. The governor’s implementation notice confirms November 5, 2025 effectiveness.

The enacted S 9008C provisions address integrated companions on defined platforms. Part Y takes effect January 1, 2027, with implementation rulemaking authorized earlier. Its age-assurance provisions specify an interim standard where the referenced rules are not yet in effect. S 9051B and S 9408A, discussed below, remain separate pending measures.

Practical duty: distinguish a standalone companion, a platform’s integrated companion, and an administrative clinical tool. Their statutory definitions differ. A crisis referral feature does not establish that a licensed person is monitoring the conversation.

What changes in Oregon and Washington on January 1, 2027?

Both states enacted companion chatbot safeguards, but their wording should not be merged into a single regional rule.

Oregon: SB 1546 addresses AI identity, self-harm protocols and additional protections for known minors, including break reminders. Practical duty: review the operator definition and the circumstances triggering the notices before incorporating a companion into between-session work.

Washington: HB 2225, Chapter 168 conditions deployment on maintaining and implementing a crisis protocol and adds identity and minor-protection provisions. Practical duty: require a concrete description of detection, response and escalation, and determine which functions your practice controls.

The Washington legislative history confirms January 1, 2027 effectiveness. The Oregon Department of Human Services session report independently confirms January 1, 2027 for SB 1546.

Which pending bills could change clinical duties next?

Pending bills do not create the duties described in this table. The current version and legislative session matter: a bill number can refer to an unrelated measure in a different session.

What should clinicians in states with pending bills do now?

California: monitor the governor’s disposition of SB 903. Its enrolled text permits specified consent verbally or in writing if documented in the record, and addresses review of therapeutic decisions, screening and other outputs. Do not use an earlier draft’s wording as the final proposal. The official status remained enrolled at review. SB 300 is on the inactive file, not an enacted extension of SB 243.

Illinois: continue to follow the enacted therapy Act. HB 5044’s liability proposal does not replace it. Track new proposals without representing a committee referral as a legal change.

Massachusetts: follow the current S 243 notice proposal if deploying a consumer-facing chatbot. The often-cited H 1974 belonged to the 2023–2024 session and accompanied a study order on June 6, 2024. It is not a current enacted requirement to obtain board approval for AI treatment. See that bill’s history.

Michigan: obtain the current SB 760 text and official history before describing its minor-access provisions as law. Meanwhile, evaluate age, consent, privacy and clinical suitability under the rules already governing your service.

New Jersey: S 735’s verified introduced text concerns professional advertising claims. A separate mental health chatbot privacy proposal was identified in secondary coverage, but its current-session identifier and disposition could not be verified. That gap must be closed before claiming complete New Jersey coverage. Do not reuse A 5603 or S 4463 from an earlier legislative session without checking whether and how they were reintroduced.

New York: distinguish current Article 47 duties from pending S 9051B and S 9408A. A legislative passage indicator is not a governor’s signature. Recheck each bill before setting a new patient-facing policy date.

North Carolina: S 963’s current history shows a committee referral, not enactment. Its proposed licensing model should inform monitoring, not be described as an existing chatbot licensing obligation.

Ohio: HB 525 proposes regulating AI therapy use; the official page shows it in House committee. Do not tell patients that Ohio has enacted this proposed ban. Complete bill-text verification remains necessary for a provision-by-provision account.

Pennsylvania: keep the clinical restriction proposal, chatbot privacy proposal, and broader companion safeguards separate. A September 3, 2026 school mental health cosponsorship memorandum expresses legislative intent; a memorandum alone is not an introduced bill or enacted law.

Which other states still need an explicit coverage check?

The following ledger prevents silence from being mistaken for clearance. It records unresolved searches and relevant inactive leads; it does not invent a “no AI law” finding. Every unresolved item in this ledger needs a current official session search before the register can be represented as exhaustive.

For further discovery, use the NCSL AI legislation database and Future of Privacy Forum chatbot tracker, then return to the official legislature. Neither a tracker category nor a search result’s proposed effective date establishes that a bill became law.

How should a clinician turn a state law into a documented decision?

Create a short decision record for each tool and use case. Identify the patient jurisdiction, professional license, product version, function, data flow and responsible person. State whether the tool records, transcribes, summarizes, generates clinical recommendations, communicates directly with patients, or makes decisions without review. Avoid describing all those activities simply as “AI assistance.”

Then identify the statutory actor and the operative provision. Ask whether the practice is a provider, supplier, deployer or operator under that law. Quote only the minimum necessary legal language in an internal record, attach the source, and record the date checked. Distinguish a legal requirement from a practice’s additional safeguard.

Document how consent is obtained and withdrawn, whether a non-AI pathway exists, how generated material is reviewed, and what happens when the tool behaves unexpectedly. If the law prohibits a function, changing the consent form is not the remedy. Disable or redesign that function.

For a fictional example, a clinic might approve a tool for scheduling after reviewing the relevant exception, decline its automated therapeutic messaging feature, and separately evaluate transcription. That is a clearer decision than approving the vendor’s entire platform under a single signature.

What should the monthly register review update?

Each review should reopen the official history and current text for every entry, resolve outstanding items, and search for new bills across all jurisdictions. Record amendments, signature or veto, chapter number, effective and operative dates, regulations, and relevant court orders. A bill that failed should move to an inactive record rather than quietly disappear.

The September 12, 2026 edition adds the newly signed California measures, distinguishes Colorado’s active psychotherapy law from future operator duties, adds Maine’s therapy Act, and separates Tennessee’s enacted studies from earlier proposed restrictions. It also preserves unresolved retrieval and coverage items rather than presenting them as verified law.

This file has no automatic monthly refresh. Comprehensive litigation monitoring remains outside the verified coverage of this edition. Before operational reliance, confirm that the displayed review date is recent enough for the decision and that the relevant entry’s verification gaps have been resolved.

Where can clinicians continue learning about AI and telehealth?

For clinical reasoning, consent, privacy, and oversight workflows, see AI in Clinical Practice: Ethics, Consent and Documentation. For jurisdiction and cross-state practice questions, see Telehealth Law and Ethics. Course approvals are in progress; these references do not claim current CE approval or satisfaction of a state’s training requirement.

You can also start with the free hour at /ce/free. Check your board’s requirements and the course’s approval information before relying on any activity for renewal credit.