Invest1 distinct publisher3 min readUpdated
The bill cleared Assembly Appropriations 13-0 and has collected one no vote across four recorded tallies. If it passes, "companion chatbot" stops being positioning and becomes a regulated category.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
California's Senate Bill 903 cleared the Assembly Appropriations Committee 13-0 on August 13 and was ordered to a third reading, leaving an Assembly floor vote, a Senate concurrence vote on the amended text, and the governor's signature between it and law [3][4][11][14]. The bill would bar companion chatbots from offering psychotherapy, restrict AI in psychotherapy to support tasks, and require clinician oversight and patient consent [1][8][9].
The commercially important detail is that the bill names the category by its marketing term. SB 903 prohibits companion chatbots from offering psychotherapy [1], which means describing a product as a companion rather than a therapist is not a defence, it is the thing being regulated. AI could still assist clinicians under the bill, but could not make independent therapeutic decisions [2]. The exemptions are narrow and specific: peer support, religious counselling, research, and federally approved medical tools [10]. The bill would also protect therapy data [9].
Read that list as a compliance map. If your product does intake, screening, or anything a reasonable regulator would call a therapeutic decision, the cheap paths out are a federal approval you probably do not have, or a clinician in the loop you probably have not staffed. Bill author State Senator Steve Padilla framed the intent plainly when introducing the measure, saying AI algorithms are not fit to take over the job of human therapists and that guardrails are needed to prevent deployment in ways that are potentially harmful to patients [5][6].
The vote arithmetic is the part operators should not talk themselves out of. The Senate passed SB 903 39-0 in May, two Assembly policy committees cleared it 17-0 and 14-1, and Appropriations went 13-0 [3][12][13]. That is 83 yes votes against one no across four recorded tallies [15]. Introduced on January 21, the bill has moved through that sequence in roughly seven months [4][16]. No date has been announced for the Assembly floor vote [7].
The opposition case is about supply, not principle. Robert Boykin, executive director of the industry group TechNet, told CalMatters that with every county in California facing a shortage of behavioural healthcare workers, SB 903 puts a clinician bottleneck in front of the intake and screening tools that help patients reach care faster [17]. That argument is real, and it is also an admission about where these tools are already being used.
Demand is not in dispute. An American Psychological Association survey in June found 77% of psychologists had patients who used AI for mental health purposes [18]. Researchers at King's College London, the Protestant University of Applied Sciences, and the City University of New York found chatbots can reinforce delusions through excessive agreement and personalised responses, describing the failure mode as an echo chamber of one where the corrective influence of real-life social interaction is absent [19][20].
Watch the Assembly floor calendar, since no date is set [7], and then the Senate concurrence vote on the amended text [14]. Watch whether the amendments widen the exemption list beyond peer support, religious counselling, research, and federally approved tools [10], because that list is the boundary of the addressable market in the largest state. And watch whether anyone shipping into California starts pricing clinician review into unit economics before they are forced to.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
SB 903 would prohibit companion chatbots from offering psychotherapy.
Under SB 903, AI could assist clinicians but could not make independent therapeutic decisions.
SB 903 cleared the California Assembly Appropriations Committee in a 13-0 vote on August 13.
Padilla said in a statement when introducing the measure: "AI algorithms are not fit to take over the job of human therapists, who have skills and training that AI is incapable of replicating... We must act and place guardrails that prevent AI algorithms from being deployed in a way that is potentially harmful to patients."
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Specific legislative record, single outlet
The claims are concrete and checkable: named bill and author, introduction date, four recorded vote tallies, an order to third reading, and direct quotes from the author and an industry critic. But everything rests on one publisher's secondary account with no statutory text, amendment detail, enforcement provisions or independent confirmation, which caps evidential strength.
Pending bill; underlying AI use already widespread
The regulation itself has no adoption yet: it has passed one chamber and committees but still needs an Assembly floor vote with no announced date, Senate concurrence and a signature, so no obligations bind anyone today. What is measurably adopted is the behavior being targeted, with a reported survey showing 77% of psychologists had patients using AI for mental health.
Headline framing broader than the reported bill
The source's headline says California 'wants it banned', while the body describes a narrower measure: chatbots may not deliver psychotherapy or pose as therapists, assistive use under clinician oversight remains allowed, and peer support, religious counseling, research and federally approved medical tools are exempt. The bill also is not yet law. The body text is measured, so the overstatement is framing rather than fabricated substance.
Named stakeholder interests on both sides
The reporting makes the interest structure legible: the bill's author frames it as patient protection, a technology trade group whose members build such tools argues clinician gating bottlenecks intake and screening, and a psychologists' association supplies the usage statistic that supports scope-of-practice restrictions. What is missing is any response from the chatbot vendors most directly exposed, and the outlet's own positioning is not disclosed.
Single-source but internally consistent
Confidence is moderate: the factual spine (votes, dates, quotes, procedural status) is consistent and specific enough to be verified against public legislative records, but with one publisher, no primary text and unspecified August amendments, the assessment of scope and consequence could shift materially with corroboration.
science
The self-driving lab is out. Whether AI shows up in your filing is still open.1 distinct publisher
invest
Bitcoin's $72,000 Break Was Mostly Forced Covering, Not Fresh Bids1 distinct publisher
leadership
Anthropic's own telemetry: 93% of permission prompts approved. Budget for blast radius, not reviewers1 distinct publisher
security
California's AI security push is really a hiring order: one AI cyber officer per agency1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 16, 2026