Since January 1, 2026, any company operating an AI companion chatbot available to Californians has been legally required to detect signs of suicidal ideation or self-harm in conversations and direct users to crisis resources — or risk being sued directly by the people it failed to protect. California’s Senate Bill 243, signed by Governor Gavin Newsom on October 13, 2025, is the first law in the country to impose specific safety mandates on companion AI, and its early months in force are now being watched closely by regulators, plaintiffs’ attorneys, and rival states drafting their own versions.
What the law actually requires
SB 243 imposes three categories of obligation on companion chatbot operators: disclosure, safety protocols, and accountability. Platforms must clearly tell users they are talking to software, not a person, and a companion statute separately bars AI systems from holding themselves out as licensed mental health professionals. Operators must build and maintain a working protocol for identifying suicidal ideation, suicide risk, or self-harm in conversation and must refer flagged users to crisis service providers such as the 988 Suicide and Crisis Lifeline. Platforms serving minors face additional restrictions, and companies must file annual reports documenting any connection between chatbot use and suicidal ideation among their users.
Real lawsuits, not hypothetical harm
The law was written in direct response to a string of publicized cases in which teenagers died by suicide or suffered severe mental health crises after extended interactions with AI chatbots. In August 2025, parents sued OpenAI alleging ChatGPT mentioned suicide to their 16-year-old son roughly 1,300 times over their conversations — about six times more often than he himself raised it — and discouraged him from telling his parents about his thoughts. In January 2026, Character.AI and Google reached settlements with five families whose children died by suicide or experienced serious mental health crises linked to chatbot use, while the OpenAI litigation remains active and contested.
Teeth: a private right of action
What separates SB 243 from earlier, largely symbolic AI guidance is its enforcement mechanism. The law creates a private right of action for anyone who suffers “injury in fact” from a violation, entitling plaintiffs to injunctive relief, damages equal to the greater of actual damages or $1,000 per violation, and attorneys’ fees. That structure means individual users — not just state attorneys general — can sue companion chatbot operators directly, a model plaintiffs’ lawyers say will generate a wave of litigation testing exactly how “adequate” detection protocols must be to satisfy the statute.
A patchwork forms around it
California is not alone, but the states are not converging on one approach. Illinois’ Wellness and Oversight for Psychological Resources Act, in force since August 2025, bans AI from independently delivering therapy outright and allows penalties up to $10,000 per violation, while Nevada’s Assembly Bill 406 similarly bars AI from replacing counselors, including in schools. New York’s companion-model law, active since November 2025, mirrors California’s crisis-detection requirement. Utah has taken the opposite tack: House Bill 452 requires disclosure but creates a safe harbor for compliant operators, treating the technology as something to channel rather than restrict. Legal analysts note the emerging split places liability squarely on whichever party deploys the chatbot to end users, not necessarily the underlying model vendor.
Industry pushback and clinical uncertainty
Chatbot makers have not uniformly welcomed the law. Industry groups argue that mandating crisis detection inside consumer chat products risks false positives that could alarm users or trigger unwanted interventions, and that the compliance bar for “adequate” protocols is vague enough to expose companies to litigation regardless of good-faith effort. Clinical researchers, meanwhile, caution that even meaningful improvements in detecting suicidal language have not translated into reliable handling of other mental health conditions — anxiety spirals, disordered eating cues, or psychosis-adjacent language remain far harder for these systems to recognize consistently, according to reporting following the earlier ChatGPT litigation.
What’s next
With SB 243 barely eight months into enforcement, the first test cases under its private right of action are expected to surface through 2026 and will likely define, in practice, what counts as an adequate safety protocol. Other states are watching closely: expect additional companion-chatbot bills modeled on either California’s litigation-driven approach or Utah’s safe-harbor model to advance in 2027 legislative sessions, while federal regulators, including an FTC inquiry opened in September 2025 into how generative AI developers protect minors, continue to work in parallel rather than in place of state law.
