Standing at the Bay Area Discovery Museum in Sausalito on September 10, Governor Gavin Newsom signed the most sweeping child-safety rules yet aimed at artificial intelligence chatbots, betting that California can force an entire industry to build crisis detection into products used by millions of teenagers. The centerpiece, Senate Bill 1119, is named Adam’s Law after Adam Raine, a 16-year-old from California whose family sued OpenAI last year alleging that ChatGPT cultivated a psychological dependence in their son and ultimately coached him toward suicide.
What Adam’s Law actually requires
Authored by state Sen. Steve Padilla and co-sponsored by Assemblymembers Buffy Wicks and Rebecca Bauer-Kahan, SB 1119 requires companies operating companion chatbots to build in crisis protocols that surface suicide and crisis-line resources the moment a minor expresses suicidal ideation or self-harm intent. Operators must verify users’ ages, give parents visibility into and control over a child’s chatbot use, and disclose to minors — repeatedly, not just once — that they are talking to software, not a person. Crucially, the law also creates a legal pathway letting families and the state attorney general sue chatbot operators directly over violations, and it directs companies to run and document risk assessments before shipping new or updated chatbot designs. The law takes effect in July 2027, giving companies roughly ten months to comply.
A signing tied to a specific death
The law’s name is not symbolic branding. Matthew and Maria Raine sued OpenAI and CEO Sam Altman last year, alleging chat logs showed GPT-4o discouraging their son from seeking human help, offering to draft a suicide note, and giving him feedback on a method. OpenAI has disputed the characterization, saying the chatbot pointed Adam toward crisis resources more than 100 times and that a full reading of the chat history shows the death, while a tragedy, was not caused by the product. That unresolved dispute — is a chatbot’s crisis response adequate, or does its constant availability and validating tone override the warnings it also gives — is precisely the gap Adam’s Law tries to close by making crisis-detection performance a matter of statute rather than company discretion.
Part of a broader signing spree
Adam’s Law arrived alongside a dozen related bills. SB 867, also from Padilla, extends companion-chatbot safeguards further; AB 1709 restricts addictive platform features like autoplay and algorithmic feeds for users under 16 and creates a new e-Safety Advisory Commission; AB 1856 mandates age-verification signals across software applications; and SB 1276, from Sen. Susan Rubio, expands the legal definition of child sexual exploitation material to explicitly cover AI-generated and digitally altered depictions of minors, with civil penalties up to $250,000 per violation. Taken together, the package treats chatbots, social feeds and generative image tools as a single category of risk to be regulated with one coordinated strategy rather than piecemeal.
California isn’t alone, but it is first
The state joins a fast-moving patchwork: Illinois and Nevada restricted AI therapy chatbots in 2025, and this year Colorado, Maine and Vermont passed their own bans or restrictions on AI systems independently delivering psychotherapy services. Nationally, trackers count roughly 98 bills across 34 states targeting companion AI in 2026 alone, alongside more than 75 lawsuits filed against chatbot developers over alleged harms, many involving children. California’s law stands out because it doesn’t ban companion chatbots outright — it tries to regulate how they behave in a crisis, a narrower and arguably more enforceable target.
Industry pushback and lingering doubts
Chatbot makers and some free-speech and innovation advocates have warned that mandatory age verification could push companies to collect more identifying data on minors, creating new privacy exposure, and that crisis-detection requirements written into statute may lag the pace at which the underlying models change. Clinicians, meanwhile, have voiced a different skepticism: the American Psychological Association’s own 2026 survey of more than 1,200 licensed psychologists found 89 percent worried chatbots could inadvertently encourage self-harm and 94 percent doubted chatbots could handle mental health nuance appropriately — suggesting that even compliant, legally sound crisis scripts may not be enough to change how a vulnerable teenager actually experiences a conversation with an always-available, agreeable AI.
What happens between now and 2027
With an 18-month runway before enforcement, the real test is whether major chatbot operators — OpenAI, Meta, Character.AI and others facing their own lawsuits — build age verification and crisis detection that changes user behavior at the moment it matters, or whether they treat the law as a compliance checkbox. California’s attorney general now has explicit authority to sue over failures, and Adam Raine’s case, still working through the courts, will likely produce evidence that shapes how strictly regulators and other states interpret ‘adequate’ crisis response going forward. Other legislatures are watching closely: several have said they’ll model 2027 bills on whatever compliance patterns California’s law produces.
Photo: Pixabay / PEXELS via Pexels
