California's bill-signing period ended on September 30, 2026. Governor Gavin Newsom enacted 12 AI-related measures and vetoed two, according to the Transparency Coalition's October 2 update. The package includes chatbot safety rules (SB 1119), customer-service chatbot disclosure (AB 1609), a toy-chatbot ban (SB 867), synthetic-performer ad disclosure (SB 1050) and a workforce-displacement notice law (SB 951). Newsom vetoed AB 2575, which would have let clinicians override AI clinical decision systems, and SB 903, which addressed AI transcription of patient mental-health information.
The package is more measured than the headline count suggests. Most of it asks companies to tell people what they are dealing with. Where California did ban something, it attached an expiry date.
The strongest case for the package
The argument for acting is serious. Companion chatbots are new, children are the likeliest to be harmed, and waiting for harm to accumulate before legislating has a poor record in tech policy. Washington has not filled the gap. The September 2026 Tech Policy Press roundup reports that Republican senators blocked attempts to fast-track two AI safety bills, leaving states and private litigants to move on their own. If Congress will not set a floor, a large state setting one is a defensible response.
Disclosure is the proportionate tool
The disclosure measures are the easiest to defend. SB 1050 requires a clear and conspicuous disclosure when an advertisement prominently includes a synthetic performer. AB 1609 requires notice when consumers interact with a customer-service chatbot. Neither stops a product from reaching market. They treat the consumer as an adult who can discount a synthetic spokesperson once told that one is on screen. That fits a pro-speech view: the state compels a narrow, verifiable fact and does not regulate the content of the message.
SB 1119 goes further. A law-firm summary of the session describes it as adding protections for minors on companion chatbots, including independent child-safety audit requirements. Audits are a heavier obligation than a label. They are still a process requirement, and they let developers decide how to meet the standard. That is a better model than prescribing design choices from Sacramento.
The toy ban is blunt but time-limited
SB 867 is the one outright prohibition. Per the bill's legislative record, it bars manufacturing, selling or offering for sale toys that incorporate companion chatbots. A toy is a product designed for play by children under 16. It was approved on September 10, 2026, and it expires on January 1, 2031.
That sunset is the feature to keep. A five-year ban on a category that barely exists is a bet that the evidence will arrive before the ban lapses. If it does, lawmakers can replace the ban with a narrower rule. If it does not, the restriction ends without further action. Pennsylvania's House held a hearing in the same week on HB 2637, a three-year moratorium on AI toys for children under 13, so California's design is likely to be copied. Legislators elsewhere should copy the sunset along with the ban.
The weakness is scope. The ban covers a whole product category, including toys with safeguards a developer could demonstrate, and it gives firms no route to show they have solved the problem. A safe-harbor or certification path would have reached the same goal at lower cost to innovation.
The workforce provision needs the closest reading
SB 951 is the hardest to assess. Early press and law-firm summaries described the pending bill as extending Cal-WARN notice to 90 days when layoffs stem from AI or automation. Fisher Phillips' analysis of the bill described that version while it was pending. The text recorded on the legislature's SB 951 page is more specific about what employers must disclose. Notices must say when a layoff results substantially from an AI system or other automated technology. They must identify the job functions being automated and the type of technology involved. The Employment Development Department must publish summaries and report to the Legislature by January 1, 2028. The page records approval on September 30, 2026 as Chapter 860. The existing 60-day notice period is unchanged, and the model is transparency plus data collection.
That is better than a hiring or firing restriction. It also carries a real risk. Attributing a layoff to AI is a legal judgment, and employers may over-report or avoid the label to limit exposure. Without careful drafting, a rule meant to measure displacement could instead produce litigation over causation.
The preemption question
The larger question is whether any of this survives a federal framework. The Tech Policy Press roundup notes that researchers opposed a still-unreleased frontier AI bill that would reportedly preempt some state AI safety laws. It also notes that attorneys general from 25 states asked Congress to preserve state authority while setting baseline federal standards. Both sides have a point. A patchwork of 50 regimes is costly for developers. A federal ceiling set in a stalemated Congress, with no floor beneath it, would leave users unprotected.
The workable settlement is narrow preemption. A federal rule should set one national disclosure standard for chatbots and synthetic media, and it should leave states free to act on child safety and product liability.
What to watch
The Transparency Coalition counts 85 new AI-related laws passed in 27 states so far in 2026. California's 12 are a fraction of that, but California sets the template others copy. Three tests will show whether this package was well designed. First, whether SB 1119's audits produce usable standards rather than paperwork. Second, whether SB 951's data shows measurable AI displacement or mostly mislabeled layoffs. Third, whether lawmakers use the 2031 sunset to evaluate the toy ban or simply extend it. A law that discloses, audits and expires gives regulators evidence, and it limits the cost to the open internet if the first draft was wrong.