Mental health professionals: artificial intelligence.

The bill's stated scope changed from regulating AI use 'by licensed professionals' to regulating AI use 'in connection with providing or facilitating psychotherapy services,' broadening who is covered beyond just licensed individuals.

New Section 4989.84(e) explicitly states that no prior review and approval by a licensed professional is required for: (1) recommending an individual discuss psychotherapy needs with a professional in response to serious mental health concerns, (2) administering standardized screening questionnaires and transmitting scores to a professional without using them to assess or advise, and (3) scheduling, confirming, canceling appointments or providing logistical information.

The exemption for AI tools reviewed and cleared by the FDA or another federal agency (former Section 4989.87(f)) was removed entirely, meaning FDA-cleared AI tools are no longer automatically exempt from the bill's requirements.

New Section 4989.84(c)(2) creates a safe harbor: a licensed professional cannot be disciplined for a violation that results solely from a feature, configuration, or function of an AI tool not within their control, if use of the tool was required by their employer or contracting entity.

Section 4989.83(a) now includes the phrase 'Except as otherwise provided in Section 4989.84,' creating a cross-reference that carves out exceptions to the general rule limiting AI to administrative and supplementary support only.

The prohibition on AI making therapeutic decisions was changed from 'independent therapeutic decisions' to simply 'therapeutic decisions,' removing the word 'independent' and thus prohibiting AI from making any therapeutic decisions without licensed professional review, not just independent ones.

New Section 4989.86(e) provides that if jurisdiction is unclear, the Director of the Department of Consumer Affairs or their designee shall determine which licensing board or enforcement agency has authority.

The definition of 'licensed professional' was revised: a licensed educational psychologist was added as an acceptable supervisor for associate professional clinical counselors, and the supervision requirements for clinical counselor trainees were simplified by removing the specific list of acceptable supervisor types.

The definition of 'licensed professional' was expanded from 7 categories to 17, adding school psychologists, registered psychological associates, associate MFTs, associate clinical social workers, psychological trainees, MFT trainees, psychiatric-mental health registered nurses with master's degrees, advanced practice RNs in psychiatric-mental health nursing, persons authorized under Family Code Section 6924, professional clinical counselors, associate professional clinical counselors, clinical counselor trainees, licensed educational psychologists, and social work interns.

New prohibition added in Section 4989.84(a) barring any individual, corporation, or entity from advertising or purporting to offer psychotherapy services when the services are provided through companion chatbots, including by claiming the chatbot is a therapist or provides therapy.

Section 4989.84(b)(2) was strengthened: AI tools that directly interact with patients in psychotherapeutic communication now must be 'approved by the United States Food and Drug Administration for that use' rather than merely being 'consistent with' FDA guidance for low-risk general wellness products or clinical support software.

Section 4989.85 data protection provision was narrowed: the Senate version flatly prohibited companies from sharing, selling, storing, or training models on psychotherapy data; the Assembly version adds 'in a manner inconsistent with any applicable law,' meaning such actions are only prohibited when they violate other existing laws.

The consent definition was changed from requiring a 'written agreement, including a written agreement provided by electronic means' to allowing 'either verbally or in writing, and documented in the record,' broadening acceptable forms of consent.

The definition of 'psychotherapeutic communication' was expanded to include interactions 'conducted for the purpose of diagnosing or treating a mental health or substance use disorder or concern,' and a new exclusion was added for 'general wellness education, instruction, or guidance that is intended to promote overall health and well-being rather than to diagnose or treat a specific mental, emotional, or behavioral health disorder or concern.'

The exception for terms of use agreements in the consent definition was narrowed: the Senate version excluded ToU agreements containing AI descriptions 'along with other unrelated information'; the Assembly version limits this to information 'unrelated to the privacy of medical information,' potentially allowing ToU agreements that mention medical privacy to qualify as valid consent.

Two new exemptions added to Section 4989.87: (e) research conducted by academic or nonprofit institutions in accordance with the Common Rule, ICH guidelines, or FDA human subject protection requirements; and (f) AI tools reviewed and cleared, authorized, or approved by the FDA or another federal agency for health care use, provided they are used consistent with their approved indication.

The fiscal note changed 'State-mandated local program' from 'no' to 'yes,' indicating the bill now imposes state-mandated costs on local agencies or school districts.

Assembly Members Addis, Lowenthal, and Pellerin were added as coauthors of the bill.

Throughout the bill, 'therapy or psychotherapy services' was shortened to 'psychotherapy services,' 'client' was added alongside 'patient' in multiple provisions, and 'therapeutic communication' was renamed to 'psychotherapeutic communication.'

The prohibition on using AI to record or transcribe sessions now applies to any individual, corporation, or entity, not just licensed professionals. A new category of 'triage or screening' was added as a regulated activity.

New exemption added: the chapter does not apply to artificial intelligence used solely for training or simulation purposes.

Consent for AI use in recording/transcribing sessions can now be given verbally (documented in the record), not only in writing as required in the introduced version.

New definition of 'triage or screening' added: assessment of an individual's health concerns and symptoms to determine urgency, clinical nature, or appropriate level of need for psychotherapy services.

The FDA standard for AI tools that directly interact with clients was relaxed from requiring FDA approval to merely being 'consistent with FDA guidance for low-risk general wellness products or clinical support software.'

New prohibition: AI may not assess an individual's health concerns or symptoms to determine urgency, clinical nature, or appropriate level of need for psychotherapy services.

New employer/employee responsibility allocation: if a licensed professional uses AI not mandated by their employer, the professional is responsible for compliance and clinical appropriateness. If the employer mandates the AI, the employer is responsible for compliance and directing use, while the professional must still use it in a clinically appropriate manner.

New data protection provision: AI use in psychotherapy records must comply with Civil Code Section 56.104 confidentiality requirements, and no company or entity may share, sell, store, or train models on data obtained from psychotherapy.

New patient rights protection: a patient does not surrender any rights to care if they or their authorized representative do not provide consent to AI use.

Enforcement shifted from Department of Consumer Affairs imposing civil penalties (up to $10,000 per violation) to the jurisdiction of appropriate health care professional licensing boards or enforcement agencies, which may pursue injunctions and restraining orders under Section 125.5 and adopt implementing rules and regulations.

A new permissive provision (Section 4989.83(a)) now explicitly authorizes individuals, corporations, or entities that provide or facilitate psychotherapy services to use AI tools for administrative support or supplementary support. The old version only contained prohibitions without an affirmative authorization.

The bill's applicability is expanded from entities that 'provide' psychotherapy services to those that 'provide or facilitate' them, potentially covering platforms and intermediaries that connect patients with therapists.

A new definition of 'companion chatbot' is added, incorporating the meaning from Section 22601 by reference, tying the bill's chatbot prohibition to an existing statutory standard.

The definition of 'licensed professional' is expanded to include (1) a psychiatric-mental health nurse practitioner certified as an advanced practice registered nurse with national certification across the lifespan, and (2) licensed educational psychologists are added to the list of supervisors for associate MFTs and MFT trainees.

The definition of 'psychotherapeutic communication' is narrowed. The old version included five subcategories covering emotional support, behavioral feedback, and other interactions. The new version reduces this to three categories focused on diagnosing/treating mental health or substance use disorders, removing broader language about emotional support and behavioral feedback.

In the 'supplementary support' definition, the requirement that data analysis be of 'anonymized data' is removed. AI may now analyze non-anonymized patient data to track progress or identify trends, as long as it is reviewed by a licensed professional.

A new definition of 'use of artificial intelligence' is added, limiting the term to AI use for administrative support or supplementary support in psychotherapy services, which frames the chapter's regulatory scope around these two categories.

The prohibition on AI performing certain acts (making therapeutic decisions, detecting emotions, generating treatment plans) is changed from an absolute ban to a requirement that these actions not occur 'without review and approval by a licensed professional.' This allows AI to perform these functions if a licensed professional reviews and approves the output.

The FDA-related exception for AI tools that directly interact with patients in psychotherapeutic communication is tightened. The old version referenced consistency with FDA guidance for 'low-risk general wellness products or clinical support software.' The new version requires the tool to be 'approved or cleared by the United States Food and Drug Administration for that use,' a higher standard.

A new standalone prohibition is added: AI shall not 'perform triage or screening' without licensed professional review and approval, in addition to the existing item about assessing health concerns.