California lawmakers have sent Governor Gavin Newsom a bill that would draw firm lines around AI in mental health care.
SB-903, authored by Senator Steve Padilla, cleared the state Senate unanimously and the Assembly 71 to 4. It keeps licensed professionals firmly in charge of psychotherapy. AI could handle administrative support or supplementary help alongside therapy, but it could not run sessions on its own, and a licensed clinician would have to oversee any tool that touches clinical judgment, from diagnoses and treatment plans to assessment results and emotional-state detection.
Consent runs through the bill. Patients, or their authorized representatives, must be told the specific purpose before AI records, transcribes, triages, or screens a session, and they must agree. Marketing a companion chatbot as a therapist would be off limits.
The push lands as AI spreads through clinical and administrative work. Supporters argue mental health is where error carries the heaviest cost, and that oversight and consent are the minimum price of adoption. Federal agencies have pushed the other way, weighing AI agents that could take part in therapy and medication decisions under Medicare.
For compliance teams, the signal is that state-level rules for AI in clinical care are no longer hypothetical.
