SACRAMENTO — Seeking to draw a sharp line between human empathy and machine learning, the California State Senate unanimously passed a bill today that would ban artificial intelligence from advertising or providing psychotherapy.
Senate Bill 903, authored by Sen. Steve Padilla (D-Chula Vista), passed on a 39-0 bipartisan vote. The legislation aims to establish guardrails around the rapidly growing market of mental health chatbots, ensuring that AI remains a tool for licensed professionals rather than a replacement for them.
The move comes as desperate patients increasingly turn to technology to fill gaps in a severely strained healthcare system. Nearly one-third of California residents live in areas with an acute shortage of mental health professionals.
According to data cited by lawmakers, therapy and companionship have emerged as some of the most popular use cases for generative AI. A wave of new tech startups has rushed to meet that demand, marketing “AI therapists” under names like Therabot, Wysa, and TherapistGPT.
But mental health advocates and clinicians warn that autonomous AI tools carry deep risks for vulnerable patients.
“As technological capacity continues to grow and new tools are introduced into the mental health space, we must ensure adequate guardrails are in place,” Padilla said. “SB 903 draws a clear line: AI can be a tool in the hands of licensed professionals, but it cannot be the professional itself.”
Risks in the machine
While tech companies pitch chatbots as affordable, 24/7 alternatives to traditional care, clinicians argue that algorithms lack the fundamental human traits required for effective therapy.
“Artificial intelligence is being implemented in our workplaces, and we’ve already seen cases where AI is recommending the wrong type of treatment for our patients,” said Ilana Marcucci-Morris, a therapist for Kaiser Permanente in Northern California and a member of the National Union of Healthcare Workers.
Medical professionals have raised a litany of concerns regarding autonomous AI therapy, including:
Diagnostic failures: An inability to detect subtle communication cues, such as changes in tone, posture, or eye contact.Clinical errors: A tendency to hallucinate data or provide incorrect, potentially dangerous treatment recommendations.Privacy vulnerabilities: Risks regarding how deeply personal, highly sensitive patient data is stored, shared, and monetized.Emotional over-reliance: The danger of patients forming unhealthy attachments to software that cannot truly comprehend their trauma or crises.
The medical community itself feels unprepared for the sudden tech shift. A recent American Psychiatric Association survey revealed that 80% of its members are moderately or very concerned that mental health professionals lack adequate training to navigate AI tools safely.
Setting the guardrails
If signed into law, SB 903 would establish strict consumer protection standards. It would explicitly prohibit tech companies from marketing automated bots as genuine “therapy.” For licensed human therapists who do use AI to assist in their practices—such as for administrative notes or data tracking—the bill mandates strict rules around patient disclosure, informed consent, and data privacy.
Critically, the legislation requires a “human-in-the-loop” framework, meaning a licensed clinician must retain ultimate oversight and veto power over any AI-generated clinical suggestions.
The bill has united a powerful coalition of state healthcare advocates, drawing co-sponsorship from the California Psychological Association, the California Association of Marriage and Family Therapists (CAMFT), the California Behavioral Health Association, and the National Union of Healthcare Workers.
“Californians deserve safeguards that protect consumers, preserve clinical judgment, and ensure that technology enhances, rather than replaces, the therapeutic relationship,” said Scott Parker, CEO of the California Psychological Association.
Joy Alafia, executive director of CAMFT, emphasized that the bill is less about stifling innovation and more about protecting people from being misled. “SB 903 is intended to protect Californians, especially vulnerable individuals, from being misled into believing AI chatbots are qualified mental health providers,” Alafia said.
The bill now heads to the California State Assembly for consideration.