Approved in the California State Assembly's Appropriations Committee on Thursday (August 13) by a 13-0 vote, the Bill 903, authored by state Senator Steve Padilla, prohibits artificial intelligence chatbots from acting as psychotherapists and from making mental health decisions without human supervision.
The proposal, named the Psychological Resources Well-Being and Supervision Act, limits the use of AI in psychotherapy to support tasks, prevents companion chatbots from presenting themselves as therapists, and requires professional supervision and patient consent. The measure also protects therapy data and creates exceptions for peer support, religious counseling, research, and federally approved medical tools.
Next steps
The senator introduced the measure on January 21. The California Senate passed it in May by a 39-0 vote. Subsequently, two Assembly committees approved it by 17-0 and 14-1 votes, and the bill moved to a third reading. There is still no date for a floor vote. If the Assembly passes the amended measure, it returns to the Senate for further review before heading to Governor Gavin Newsom.
When introducing the measure, Padilla said that “AI algorithms are not equipped to take on the work of human therapists, who have skills and training that AI is incapable of replicating.” He defended the need to “put safeguards in place to prevent AI algorithms from being used in ways potentially harmful to patients.”
Criticism and context
Robert Boykin, executive director of the technology group TechNet, told CalMatters that at a time when all counties in California face a shortage of behavioral health workers, the bill “still puts a bottleneck of professionals in front of the screening tools that help patients get to care more quickly.”
The discussion comes amid concerns about the use of AI in mental health. Research by the American Psychological Association, released in June, found that 77% of psychologists treated patients who used AI for mental health purposes. Researchers from King's College London, the Protestant University of Applied Sciences, and the City University of New York concluded that chatbots can reinforce delusions through excessive agreement and personalized responses.
“The tendency of AI chatbots to agree with users' opinions has been compared to social media echo chambers and, in its most extreme form, to an ‘echo chamber of one,’ in which the positive corrective influence of real social interactions is absent,” the researchers wrote.
