© Copyright 2025 MarriageFamilyTherapist.org
The veto leaves existing duties intact. Here is how to practice and train wisely.
One approach requires licensed review and approval of AI decisions about a client's emotional state. The other leaves AI screening and documentation inside existing supervision and consent rules. On October 1, 2026, the final day to act on bills, Gov. Gavin Newsom chose the second path, vetoing SB 903, the AI therapy bill backed by CAMFT, according to the Sacramento Bee report.
The veto adds no new AI-specific mandates for LMFTs. It also does not remove existing duties around supervision, consent, confidentiality, and truthful advertising. California LMFTs must apply pre-AI oversight rules to tools that evolve faster than statute.
Senate Bill 903 was California's attempt to draw a bright line between licensed psychotherapy and unlicensed AI-driven tools. Authored by State Senator Steve Padilla and last amended on July 2, 2026,2 the measure had two central provisions. First, a therapist could not use AI to make decisions or evaluate a person's emotional state without the review and approval of a licensed professional, such as an LMFT in California.1 Second, companies and individuals could not advertise or offer chatbot-delivered services as therapy unless the care was provided by a licensed professional.
The final text allowed AI for administrative support and limited supplementary support such as documentation and workflow help within MFT practice management software, as long as a licensed professional kept responsibility for clinical decisions.2 Independent therapeutic decisions by AI were prohibited. AI could not directly interact with patients in psychotherapy unless the tool was FDA-approved for that use and HIPAA-compliant.3 The bill also required verbal or written notice before AI was used to record or transcribe psychotherapy sessions or to triage or screen clients.3
Gov. Gavin Newsom vetoed the bill on October 1, 2026, the final day for action.4 He called it "overly broad" and said it "would drastically limit a clinician's use of tools that benefit the delivery of care today, including by requiring routine screening determinations to receive direct approval."4 In plain terms, he was concerned that even routine AI-assisted screening would need a licensed clinician's sign-off before it could be used.
The bill passed the Senate unanimously and drew only one no vote in the Assembly. Padilla said the veto allowed "unlicensed algorithms to act as therapists." The Sacramento Bee reported the veto, and the final amended bill defined artificial intelligence and gave enforcement authority to California's healing arts boards.1
California's AI therapy debate has shifted from whether to regulate these tools to how narrowly any future rules should be drawn. On October 1, 2026, Gov. Gavin Newsom vetoed SB 903, so the bill does not add a new AI-specific compliance layer to California LMFT licensing.
A vetoed bill is not law. That means the state is not newly requiring a licensed clinician to approve every AI-assisted screening, documentation choice, or treatment-planning suggestion before it is used. LMFTs should not read the veto as a green light to delegate clinical judgment to a chatbot, but they also should not assume a new sign-off rule now exists for routine AI tools.
The veto does not remove existing legal and ethical duties around licensed supervision, informed consent, confidentiality, and truthful advertising. If a trainee or associate uses an AI note writer or intake tool, the supervising LMFT's review obligations remain whatever they already were. Likewise, clients still need accurate information about whether a service marketed as therapy involves a licensed professional. Newsom's "overly broad" objection was specifically that the bill would have swept in routine screening determinations and administrative tools that support care, not just chatbot-delivered therapy. That distinction is likely to shape any follow-up proposal. For verified state-law specifics, see the compliance and advertising sections.
After Gov. Gavin Newsom vetoed SB 903 on Oct. 1, 2026, California LMFTs did not receive a single new AI code, but they also did not lose the existing licensing and consumer-protection duties that already apply. The table below separates the rejected SB 903 provisions from the baseline rules clinicians must continue to follow.
| SB 903 proposed requirement | What remains after the veto |
|---|---|
| Required a licensed professional to review and approve AI tools that make decisions or evaluate a client's emotional state. | No new review mandate applies, but general supervision and standard-of-care duties remain under California law. |
| Prohibited providing psychotherapy through AI unless conducted by a licensed health-care professional. | Existing California law still limits psychotherapy practice to licensed providers, but without SB 903's AI-specific wording. |
| Barred advertising therapy services delivered by a chatbot. | Current advertising rules still prohibit false or misleading statements about licensure and services, but no chatbot-specific advertising ban exists. |
| Allowed AI only for limited administrative or supplementary support, required disclosure plus affirmative consent before recording or transcribing therapy sessions, and prohibited AI from independently interacting with clients, making therapeutic decisions, detecting emotions, or generating treatment plans without professional review. | Clinicians remain bound by California confidentiality, informed consent, and unlicensed practice rules, but these are not the proposed SB 903 AI guardrails. |
| Tied delivery of psychotherapy to professionals licensed and regulated by state licensing boards. | LMFTs and associates remain under BBS oversight through existing statutes and regulations, not through SB 903. |
California is an all-party consent state. Under Penal Code Section 632, recording confidential communications without consent can lead to fines up to $2,500 per violation and up to one year in jail. An ambient AI scribe that captures session audio is a recording, and there is no software exception. Before using any listening or transcription tool, get consent from every participant. That consent should identify the AI tool, its purpose, what audio or text it captures, how long data is retained, who can access it, and how a client can decline or withdraw. Stop capture immediately if consent is withdrawn, and document the consent in the clinical record.
Session audio, transcripts, and AI summaries can be protected health information. If a vendor can access PHI, a HIPAA business associate agreement is required. CAMFT advises reviewing that agreement carefully; choosing HIPAA compliant teletherapy platforms for MFTs does not replace that review, and a vendor's claim of compliance is not enough. For California clients, the Confidentiality of Medical Information Act also treats identifiable AI-generated records as confidential medical information, and CMIA includes a private right of action. Check whether the vendor retains raw audio, shares data across customers, trains models on client content, or uses data beyond treatment documentation. Model training or broader secondary use should require separate authorization from the client.
AI output is a draft, not a clinical judgment. A licensed clinician remains responsible for intake, screening, and triage decisions. Review each AI note for accuracy and completeness before it becomes part of the record. If the tool flags risk or suggests a diagnosis, treat that as administrative support, not a substitute for your own assessment.
Do California therapists have to tell clients they use AI? Yes, consent is required before recording or transcribing, and BBS guidance permits verbal or written disclosure while CAMFT expects AI consent as a baseline. The law is less settled on whether that disclosure must be written. For clinical documentation only, a verbal informed consent that is documented may be professionally acceptable, but written consent is safer when a vendor retains audio, uses cloud processing, transfers data outside the practice, or uses information beyond therapy documentation. Standard telehealth regulations for couples therapy do not automatically cover AI scribe use, so add a separate AI-specific disclosure.
Starting April 1, 2026, every California LMFT advertisement must include the licensee's first and last name, license number, and a complete title or allowable abbreviation (Licensed Marriage and Family Therapist, MFT, or LMFT).1
SB 903's proposed ban on advertising chatbot-delivered services as therapy did not become law. Existing California advertising and unlicensed-practice rules still apply.
Business and Professions Code Section 651 bars false, fraudulent, misleading, or deceptive public communications. The BBS treats misrepresenting license type or status, education, qualifications, or affiliations as unprofessional advertising.4 Fictitious business names cannot mislead, and patients must be told the practice owner's name and license designation before treatment, a disclosure covered in any LMFT private practice guide for California clinicians.3
Use only Licensed Marriage and Family Therapist, MFT, or LMFT for a licensed clinician.5 Associates use Registered Associate Marriage and Family Therapist; AMFT is acceptable only if the full title also appears, a distinction outlined in LMFT vs. AMFT vs. LMFT-A.5 Using MFT in an associate email address can imply licensure and be misleading.6
State who is licensed and who reviews AI output. Examples: "AI-generated intake notes reviewed by a Licensed Marriage and Family Therapist" or "automated screening tool, results reviewed by your LMFT." Avoid "AI therapy" labels that imply the chatbot itself is delivering care.
Not fully settled for every product. AB 489, effective January 1, 2026, bars using AI to provide health advice or services while representing, or letting consumers understand, that a licensed professional is delivering them when that is not true.7 BPC Sections 4999.8 and 4999.9 prohibit AI from presenting as a licensed person or using protected titles.6 The BBS says psychotherapy must be conducted by a licensed professional.8 A chatbot cannot hold itself out as an LMFT or imply licensed delivery, but a single categorical ban on every use of "therapy" is not established.
The SB 903 fight split California's mental health field along a clear line: clinician groups wanted AI-specific guardrails, while hospital and technology interests argued those rules would slow useful tools.
CAMFT co-sponsored the measure with the California Psychological Association, CBHA, and NUHW.3 CAMFT's SB 903 action page urged members to ask Gov. Newsom to sign. National Union of Healthcare Workers President Emeritus Sal Rosselli argued that "Californians still have no assurance that a product marketed as providing therapy involves a licensed professional" as chatbot mental health products spread. CBHA called the bill a way to advance AI accountability and stop algorithms from advertising mental health services without provider responsibility.2
The California Hospital Association, along with business and technology trade groups and major medical organizations, opposed the bill, saying it would interfere with AI that supports therapists. The governor's veto message echoed that concern, calling the bill overly broad.
The Board of Behavioral Sciences supported SB 903 in its SB 903 bill analysis, but as of October 2026 the board has not published a stand-alone AI practice bulletin or FAQ for LMFTs. Its analysis described AI as limited to administrative or supplementary support unless licensed review and client consent requirements are met.1 That support does not yet translate into final operational guidance for California clinicians.
For California MFT programs, trainees, and associates, the SB 903 veto does not create an AI loophole. Existing supervision rules still make the supervisor responsible for the extent, kind, and quality of a trainee's counseling, and that responsibility extends to AI-assisted decisions.3 California still requires one unit of supervision per week in each practice setting, plus one additional unit for every five hours of direct clinical counseling.1
Use simple documentation habits: record which AI tool was used, what the trainee or associate reviewed or changed, and who approved the final clinical decision. Keep that note in the supervision record, not just the client file, so the review is visible to the supervisor.
Programs and supervisors should settle in writing whether AI may be used for assessment, treatment planning, or screening, grounding that decision in each clinician's MFT scope of competence. The BBS has not published MFT-specific AI guidance, so silence should not be read as permission. A supervision agreement must be documented within 60 days and a responsibility statement signed before supervision begins.2
Don't wait for legislation. Faculty and site supervisors should adopt a written AI-use policy that defines allowable tools, required review steps, and documentation for trainees. That protects clients and gives supervisors a defensible supervision record.
