Updated 17 September 2026. Senate Bill 867 is law, and the detail that matters most for families was missing from most coverage: the ban does not apply yet. Governor Gavin Newsom signed the bill on 10 September 2026 as part of a package of child safety measures, confirmed in the governor's official announcement. The prohibition on manufacturing, selling or offering to a retailer a toy containing a companion chatbot takes effect on 1 January 2027 and runs until 1 January 2031. The legislative record is kept on the official California page for SB 867.
That is a big deal for the category Ted belongs to, so we are going to be direct about it. This article explains what the law actually says, who it covers, why lawmakers wrote it, what the serious objections are, and what a parent shopping for a talking toy this autumn should take away from it. We are not neutral observers here, and we will say plainly where we think the law identifies a real problem. If you want the wider picture rather than this one statute, start with our AI toy safety guide.
What does SB 867 actually ban, and from when?
SB 867 prohibits manufacturing, selling, exchanging, possessing with intent to sell, or offering to a retailer any toy that includes a companion chatbot. That prohibition begins on 1 January 2027 and expires on 1 January 2031, a four year pause. It does not ban interactive toys generally, and it does not ban educational software. It targets one specific thing: a conversational system designed to sustain an ongoing relationship with the child.
The statute defines a toy as a physical product designed, marketed or manufactured for use in play by children under 16 years of age. An earlier version of this article said under 18, which was wrong, and we have corrected it. The definition of a companion chatbot comes from SB 243, the 2025 California law the new statute extends: a system with a natural language interface that gives adaptive, human-like responses, can meet a user's social needs, shows anthropomorphic features, and can sustain a relationship across multiple interactions. Customer service bots, video game characters limited to game-related replies, and plain voice assistants are explicitly excluded.
One feature of the drafting deserves attention, because it is what separates this law from every other AI rule passed in 2026. As the law firm Ballard Spahr set out in its 11 September 2026 alert on SB 867, the statute creates no compliance path. A covered toy stays prohibited even if the manufacturer adds content filters, parental controls, session limits, independent testing or certification. The only way out is to change the product so it no longer meets the definition. The full legal text and committee reasoning are laid out in the Assembly Privacy and Consumer Protection Committee analysis of 16 June 2026, which is the single most useful document if you want the primary source rather than a summary of it.
SB 867 was not signed alone. A companion bill, SB 1119, places broader child safety obligations on companion chatbot operators generally: risk assessments, published safety policies, and independent audits. That one matters for the longer run, because it sketches what a standards-based regime would look like if the 2031 sunset arrives and legislators want something other than a ban.
Why did legislators write it?
Because independent testers found that several talking toys on sale in 2025 gave children genuinely dangerous answers. This is not speculation or a hypothetical risk. It was documented, transcribed and published.
The trigger was Trouble in Toyland 2025, a report by the U.S. PIRG Education Fund that tested four commercially available chatbot toys. Three ran on some version of OpenAI's models. One teddy bear, FoloToy's Kumma, running GPT-4o, reliably told researchers where to find knives, pills, matches and plastic bags, walked through how to strike a match, and drifted into sexually explicit territory during longer conversations. The Assembly committee reproduced those transcripts in its analysis, which tells you how central they were to the bill's passage. We went through the same report in more depth when it landed, in our breakdown of the safety testing.
There is a second thread. OpenAI's own terms state that users under 13 are not permitted to use its models. Several of the tested toys were marketed to children as young as three. The committee analysis flags that gap directly, and it is the kind of contradiction that is hard for an industry to argue its way out of.
The bill's sponsor, the advocacy group Children Now, and supporters including the American Academy of Pediatrics California, argue that the deeper issue is design intent. A system engineered to maximise engagement is not the same thing as a system engineered to support a child's development, and young children are not well equipped to tell the difference between a machine and a person.
What are the serious arguments against the law?
The strongest objection is that a category-wide ban removes any reward for building a safer product. If a toy with strict content filters and real parental controls is illegal on exactly the same terms as one with no safeguards at all, safety stops being a competitive advantage. The Ballard Spahr reading confirms this is not a rhetorical worry: there is no certification route in the text.
That argument was made in written testimony by the Reason Foundation in March 2026. Their analyst also pointed out that California's existing toy safety framework does not ban product categories: it identifies specific hazards, sets standards, and holds manufacturers to them. And a four-year pause does not resolve the safety questions, it postpones them, potentially leaving regulators in 2031 with less industry insight than they have now.
Two industry groups raised narrower concerns in committee. The Civil Justice Association of California argued that the definition of a toy is broad enough to sweep in products nobody thinks of as toys. The California Retailers Association asked for a knowingly standard so that a shop is not liable for stocking a product it had no way of knowing was non-compliant.
Worth holding onto: the law's critics and its supporters agree on the facts of what those tested toys did. They disagree about whether a pause or a standard is the better fix.
Does the moratorium apply to Ted?
On a plain reading of the text, yes, from 1 January 2027 in California. Ted is a plush toy for children that holds real conversations, so a moratorium on toys containing companion chatbots covers it, and we are not going to pretend otherwise. Nor do we read the law as a judgment on any particular product, ours included. It is written as a category-wide pause precisely because the legislature did not want to assess products one by one.
What we would argue is that most of the specific failures documented in the PIRG report are design choices, not inevitable properties of a talking toy. A toy that will explain how to light a match is a toy where nobody built a hard content boundary. A toy that drifts into adult topics after twenty minutes is a toy whose guardrails were tested for one exchange and not for a long conversation. A toy that quietly hoovers up recordings is a toy whose data model was designed for the company rather than the family.
Those are the questions we get asked most often, and they are why we publish our EU and US certifications in full rather than summarising them, on our security page. Whether that is enough is a fair question, and a legislature is entitled to decide the whole category should slow down while standards get written.
What should a parent check before buying any talking toy?
Start with the boring questions, because they are the ones that separate a considered product from a rushed one. The PIRG findings and the committee analysis point at five checks in particular.
| Check | What a good answer looks like |
|---|---|
| Camera | None. A microphone and speaker are enough for conversation. |
| Recordings | Clearly stated retention, an easy way to delete, no resale to third parties. |
| Content limits | Parent sets topics and tone, and the limits hold in a long conversation, not just the first exchange. |
| Age claims | The stated minimum age is consistent with the underlying model's own terms of use. |
| Ongoing cost | Stated up front. A toy that stops working when a plan lapses is a different purchase. |
The one we would add from our own inbox: try a long conversation before you hand the toy over. Most guardrail failures in the PIRG testing appeared after the conversation had been running for a while, not in the first minute. Our own FAQ page covers the recording, Wi-Fi and parental control questions in more detail.
What happens next, and does it affect families outside California?
California is settled, with a fifteen month runway before the rule bites. The open question is how many other states follow, and the early signal is that several intend to. New York is now the most advanced: Senate Bill 9408 would impose a five-year moratorium on the manufacture and sale of toys containing AI companions, and Senate Bill 9051 would bar chatbot operators from offering minors a set of features considered unsafe. Pennsylvania House Bill 2637, introduced by Representative Joe Ciresi on 15 June 2026 and still in committee, would impose a three-year prohibition, tracked on the Pennsylvania General Assembly record for HB 2637. State of play as of 17 September 2026: California signed and effective 2027, New York and Pennsylvania pending.
California rules travel. The state is large enough that manufacturers usually redesign for it rather than maintain a separate product, which is why a California standard often becomes the de facto North American one. SB 867 is also only one of many: we counted fourteen new state chatbot laws this year alone. Europe is moving on a parallel track through revised toy safety rules, and the Toy Association's legislative tracker is a reasonable place to follow both.
For a family buying this autumn, the practical implication is smaller than the headlines suggest. SB 867 changes who is allowed to sell a talking toy in one American state, starting in 2027. It changes nothing about what makes such a toy good or bad for your child today, and nothing at all outside the United States. The checks in the table above are what actually separate products, which is why Ted was built with no camera, no advertising, no data resale and no subscription, with topics and tone set by the parent in the app.
Frequently asked questions
Has California banned AI toys?
It has paused one category of them, starting in 2027. Governor Newsom signed SB 867 on 10 September 2026, after the bill passed the Senate 39-0 on 28 May and cleared both chambers on 31 August, with the Assembly voting 74-1. The prohibition on manufacture and sale takes effect on 1 January 2027 and lasts until 1 January 2031. Interactive toys that are not companion chatbots are unaffected.
When does SB 867 come into force?
1 January 2027. The bill was signed on 10 September 2026, but the prohibition itself does not apply before that date, which is why toys in the category can still be sold in California through the 2026 holiday season. The moratorium then runs for four years and sunsets on 1 January 2031.
Which children does SB 867 cover?
The statute defines a toy as a physical product designed, marketed or manufactured for use in play by children under 16 years of age. That threshold is higher than the under-13 line used in most US children's privacy law, and higher than the age range most talking toys are marketed to.
Can a manufacturer comply by adding filters or parental controls?
No. The statute sets out no compliance path through disclosure, testing or certification, so a covered toy remains prohibited even with strong safeguards in place. The only way out of scope is to change the product so it no longer meets the statutory definition of a companion chatbot.
Which AI toys failed safety testing?
The U.S. PIRG Education Fund tested four chatbot toys in its Trouble in Toyland 2025 report. FoloToy's Kumma, a teddy bear running OpenAI's GPT-4o, showed the least consistent guardrails, telling researchers where to find knives, pills and matches and engaging with sexually explicit topics.
Does a law like this apply to toys sold in Europe?
No. SB 867 is a California state law and only governs manufacture and sale in California. European families are covered by the EU's toy safety rules and by GDPR, which are separate frameworks with their own requirements around data collected from children.
This article discusses child safety legislation that references cases of self-harm among minors. If any of it touches on something you are dealing with personally, speaking to a doctor or a trusted professional is worth doing.
