As the California Legislature’s 2025 session draws to a close, lawmakers have advanced over a dozen AI bills to the final stages of the legislative process, setting the stage for a potential showdown with Governor Gavin Newsom (D).  The AI bills, some of which have already passed both chambers, reflect recent trends in state AI regulation nationwide, including AI consumer protection frameworks, guardrails for the use of AI in employment and healthcare, frontier model safety requirements, and chatbot safeguards. 

AI Consumer Protection.  California lawmakers are advancing several bills that would impose disclosure, testing, documentation, and other governance requirements for AI systems used to make or assist in decisions that impact consumers.  Like 2024’s Colorado AI Act, California’s Automated Decisions Safety Act (AB 1018) would adopt a cross-sector approach, imposing duties and requirements on developers and deployers of “automated decision systems” (“ADS”) used to make or facilitate employment, education, housing, healthcare, or other “consequential decisions” affecting natural persons.  The bill would require ADS developers and deployers to conduct impact assessments and third-party audits and comply with various disclosure and documentation requirements, and would establish consumer notice, correction, and appeal rights. 

Employment and Healthcare.  SB 7 would establish worker notice, access, and correction rights, prohibited uses, and human oversight requirements for employers that use ADS for employment-related decisions.  Other bills would impose similar restrictions on AI used in healthcare contexts.  AB 489, which passed both chambers on September 8, would prohibit representations that indicate that an AI system possesses a healthcare license or can provide professional healthcare advice.

Frontier Model Safety.  Following the 2024 passage—and Governor Newsom’s subsequent veto—of the Safe & Secure Innovation for Frontier AI Models Act (SB 1047), State Senator Scott Wiener (D-San Francisco) has led a renewed push for frontier model safety with his Transparency in Frontier AI Act (SB 53).  SB 53 would require large developers of frontier models to implement and publish a “frontier AI framework” to mitigate potential public safety harms arising from frontier model development, in addition to transparency reports and incident reporting requirements.  Unlike SB 1047, SB 53 would not require developers to implement a “full shutdown” capability for frontier models, conduct third-party audits, or meet a duty of reasonable care to prevent public safety harms.  Moreover, while SB 1047 would have established civil penalties of up to 10 percent of the cost of computing power used to train any developer’s frontier model, SB 53 would establish a uniform penalty of up to $1 million per violation of any of its frontier AI transparency provisions and would only apply to developers with annual revenues above $500 million.  Although its likelihood of passage remains uncertain, SB 53 builds on several recent state efforts to establish frontier model safeguards, including the passage of the Responsible AI Safety & Education (“RAISE”) Act in New York in May and the release of a final report on frontier AI policy by California’s Frontier AI Working Group in June.

Chatbots.  Various other California bills would establish safeguards for individuals, and particularly children, that interact with AI chatbots or generative AI systems.  The Leading Ethical AI Development (“LEAD”) for Kids Act (AB 1064), which passed the Senate on September 10 and could receive a vote in the Assembly as soon as this week, would prohibit individuals or businesses from providing “companion chatbots”—generative AI systems that simulate sustained humanlike relationships through personalization, unprompted questions, and ongoing dialogue with users—to children if the companion chatbot is “foreseeably capable” of engaging in certain activities, including encouraging a child to engage in self-harm, violence, or illegal activity, offering unlicensed mental health therapy to a child, or prioritizing user validation and engagement over child safety, among other prohibited capabilities. Another AI chatbot safety bill, SB 243, passed the Assembly on September 10 and awaits final passage in the Senate.  SB 243 would require companion chatbot operators to issue recurring disclosures to minor users, implement protocols to prevent the generation of content related to suicide or self-harm, and disclose companion chatbot protocols and other information to the state.  

The bills above reflect only some of the AI legislation pending before California lawmakers ahead of their September 12 deadline for passage.  Other AI bills have already passed both chambers and now head to the Governor, including AB 316, which would prohibit AI developers or deployers from asserting that AI “autonomously” caused harm as a legal defense, and California SB 524, which would establish restrictions on the use of AI by law enforcement agencies.  Governor Newsom will have until October 12 to sign or veto these and any other AI bills that reach his desk.