California Lawmakers Move to Ban AI From Acting as a Therapist

California is drawing a line around how far AI can go in pretending to be a therapist.

Senate Bill 903 would restrict AI firms from marketing chatbots as licensed therapy, block automated systems from making unguided clinical decisions, and require patient consent before healthcare providers use AI for tasks including recording therapy sessions or triaging care. The measure, introduced by state Sen. Steve Padilla, has already cleared the California Senate.

The debate is no longer simply whether AI belongs in mental healthcare. California lawmakers are confronting a harder question: which decisions still require a licensed human professional when chatbots can increasingly sound like one?

California draws the line

California state Sen. Steve Padilla introduced SB 903 in January and has drawn support from the National Union of Healthcare Workers.

Per The Independent, Padilla noted that AI could improve Californians’ lives if used responsibly, but also argued that “AI algorithms are not fit to take over the job of human therapists, who have skills and training that AI is incapable of replicating.”

But the proposal clarifies that it is not a general ban on AI in mental healthcare. Its focus is on AI being presented as a therapist and on clinical uses where the system could make decisions without a licensed professional involved.

The proposal has also drawn opposition from technology industry groups that argue it could restrict legitimate clinical uses of AI.

Robert Boykin, California executive director of TechNet, for instance, called the bill a “clinical bottleneck,” arguing that California is already facing a shortage of behavioral healthcare workers and that AI tools are filling that gap.

Why lawmakers say AI therapy poses risks

The American Psychological Association found that patients are increasingly using AI for mental-health support, self-diagnosis, and treatment assistance, but psychologists remain concerned that chatbots lack the clinical validation, safeguards, and human judgment needed to replace qualified professionals.

The scale of those interactions can be significant. Last October, OpenAI said around 1.2 million users each week sent ChatGPT messages containing “explicit indicators of potential suicidal planning or intent.”

The risks have also moved into the courts. The makers of several popular AI applications, such as Google and Character.AI, have faced lawsuits alleging that their chatbots played a role in youth suicides, adding a legal dimension to concerns that systems designed to sustain highly human-like conversations can respond poorly when users are in serious distress.

That does not, however, establish that a chatbot caused those deaths; the allegations remain claims in litigation. But for California lawmakers, the cases add context to a broader problem.

People are already treating conversational AI tools as mental-health support experts because they sound like one, even though they are not. The lack of consistent standards for evaluating chatbot responses has fueled concerns about relying on general-purpose AI systems for mental-health guidance, particularly during crises.

May be the first of many

California’s move could give other governments another AI policy to study as they work out where the technology should be subject to legal limits.

SB 903 could become an early test of how governments regulate AI when conversational ability begins to resemble professional expertise.

The larger question extends well beyond therapy. If lawmakers decide that some roles require enforceable boundaries between AI assistance and human professional judgment, healthcare may be only the beginning.

Other News: Australia is moving ahead with a world-first ban on social media accounts for children under 16, requiring major platforms to block or remove underage users or face fines of up to AU$49.5 million.