California AI Therapy Bill Moves Toward Assembly Vote

California's effort to keep AI chatbots from presenting themselves as therapists moved toward an Assembly vote after Senate Bill 903 cleared the Assembly Appropriations Committee 13-0, Decrypt reported, setting up a test of how far states will go in separating mental-health support tools from clinical care.

The bill, introduced on January 21, has been ordered to a third reading in the Assembly.

No Assembly vote date has been announced, and an amended version would still need Senate concurrence before reaching Governor Gavin Newsom.

SB 903, also called the Wellness and Oversight for Psychological Resources Act, would limit AI use in psychotherapy to support tasks.

Chatbots could not pose as therapists, make mental-health decisions without human oversight, or operate without clinician involvement and patient consent.

California State Senator Steve Padilla, the bill's author, framed the measure as a guardrail against replacing licensed care.

Padilla said in a statement when the measure was introduced that AI systems cannot replicate the skills and training of human therapists.

The voting record shows broad legislative support so far.

The Senate passed the bill 39-0 in May, and two Assembly policy committees later advanced it by votes of 17-0 and 14-1 before the Appropriations Committee sent it forward on Thursday.

Industry opposition centers on access and staffing.

Robert Boykin, executive director of technology industry group TechNet, warned that mandatory clinician involvement could slow intake and screening tools in counties already facing behavioral-health workforce shortages.

The policy debate is arriving as AI systems become part of ordinary mental-health behavior rather than a fringe experiment.

An American Psychological Association survey in June found that 77% of psychologists had patients who used AI for mental-health purposes.

A separate research team from universities in London, Germany and New York also warned that chatbot agreement patterns can intensify isolation.

Their paper compared sycophantic AI behavior with social-media echo chambers and, in the most extreme version, an "echo chamber of one" without corrective real-life interaction.

For AI developers and digital-health companies, the California bill would draw a bright line between administrative or support automation and treatment authority.

Intake, triage and educational functions could remain possible only where the workflow keeps a human clinician and a consenting patient inside the loop.

The next procedural step belongs to the Assembly calendar.

Until that vote is scheduled, the bill's practical effect remains a proposed boundary rather than an enforceable rule for chatbots used in California mental-health settings.