California SB 867 Kills AI Chatbot Toys for Kids 2026

The same large language models that companies are racing to embed in everything just hit a wall they did not expect: a child's toy box. California's Senate voted 39 to 0 to ban AI companion chatbots in toys for kids twelve and under, the first time any American legislature has drawn a hard line around where conversational AI may not go. The unanimity is the story. In a state that fights over almost everything, from water rights to housing to tech regulation itself, not a single senator was willing to stand up and defend conversational AI toys for small children. When a legislature this divided agrees on anything by a shutout, it signals that the underlying worry has moved from partisan talking point to settled common sense.

What Actually Happened

Senate Bill 867, authored by state senator Steve Padilla, passed the California Senate by a 39 to 0 vote and now moves to the Assembly. The bill imposes a four-year moratorium, running until January 1, 2031, on the manufacture and sale of toys equipped with generative AI companion chatbots intended for children aged twelve and under. It is not a permanent prohibition. It is a deliberate pause, written to stop a product category from scaling into millions of homes before anyone understands what it does to developing minds.

The bill is precise about its target. It defines a companion chatbot as a natural language interface capable of providing adaptive, human-like responses designed to meet a user's social or emotional needs. That definition deliberately excludes the talking dolls and pull-string toys of previous decades, which follow pre-recorded scripts. The new generation is different in kind: powered by large language models, these toys sustain long-term, evolving conversations, remember what a child told them yesterday, and adapt their personality to keep the child engaged. SB 867 treats that capacity for open-ended emotional bonding as the hazard, not the novelty.

Padilla is not a newcomer to this fight. He authored what is widely described as the nation's first chatbot safety protections, and SB 867 extends that work into the physical-product market specifically. The motivating case hangs over the entire debate: in 2024, a fourteen-year-old named Sewell Setzer III took his own life after forming an intense emotional attachment to an AI chatbot on Character.AI, having exchanged more than 1,000 messages with the bot, many of them deeply emotional or romantic. Legislators repeatedly invoked that tragedy as proof that the technology can reach a child's interior life in ways no toy ever has.

Why This Matters More Than People Think

Most coverage framed this as a narrow consumer-safety bill about a niche product. That misreads it. SB 867 is the first concrete answer to a question the entire AI industry has avoided: are there contexts where conversational AI simply should not be deployed, regardless of how good the safeguards are? The reason.org testimony opposing the bill noted that it bans all such toys regardless of what safety measures the manufacturer implements. That is the radical part. The legislature decided the risk is structural, not a matter of better guardrails, and chose prohibition over regulation.

For the toy industry, the timing is brutal. Major manufacturers have spent the last eighteen months racing to put AI companions into their flagship products, betting that a toy that talks back and remembers is the upgrade that justifies a premium price. Mattel signed a partnership with OpenAI in 2025 to explore exactly this category. A four-year California moratorium does not just close one state market; because California is so large, it effectively sets the national product roadmap. Few manufacturers will build a separate non-AI version for one state and an AI version for the rest, so the conservative move is to pull the feature everywhere.

The deeper significance is jurisdictional. AI regulation in the United States has stalled at the federal level, and the most consequential rules are now being written state by state, starting with California. SB 867 demonstrates that states will not wait for Washington, and that they will regulate AI not as an abstract technology but use case by use case: children's toys today, companion apps and elder-care bots tomorrow. That fragmented, application-specific approach is the opposite of the single national framework the industry has lobbied for, and it is becoming the actual regulatory reality.

There is a market consequence that compounds the regulatory one. Once a category carries a state ban, liability insurers and retailers grow wary long before any law takes full effect. A toy buyer at a major chain has to weigh the reputational and legal exposure of stocking a product that one large state has formally judged unsafe for children. That chilling effect moves faster than legislation and reaches further, because procurement decisions made in the next buying season will shape what appears on shelves for the holiday after. The ban's real teeth may be commercial caution, not the statute itself.

The Competitive Landscape

The companies most exposed are the ones that moved fastest. Mattel and Hasbro have both signaled AI ambitions, and a wave of venture-backed startups has launched plush toys and robots pitched explicitly as a child's AI friend. Character.AI, whose product sits at the center of the Setzer case, has become the cautionary name every legislator cites. OpenAI, through its Mattel partnership, is suddenly a stakeholder in a category that just got a four-year stop sign in its home state. The firms that hesitated on AI toys now look prudent rather than slow.

California is also not acting in a vacuum. The bill sits alongside a broader push of state-level AI legislation, and other states tend to follow California's lead on consumer protection the way they once followed its emissions and privacy standards. Padilla's earlier chatbot-protection law already established the template. If SB 867 clears the Assembly and is signed, expect copycat bills in New York, Illinois, and elsewhere within a single legislative cycle, the same diffusion pattern that turned California's privacy rules into a de facto national standard.

The historical parallel is the lawn-dart ban of the late 1980s, when the Consumer Product Safety Commission prohibited a popular toy outright after it caused child deaths, overriding the argument that careful use made it safe. The lesson then was that some products aimed at children carry risks that no instruction label can offset, and the law will remove them from the market entirely rather than manage the risk. SB 867 applies that same logic to software: when the hazard is the open-ended emotional relationship itself, a warning sticker and a content filter are not enough. The category, not the configuration, is the problem.

The contrast with the social-media era is instructive. When Instagram and TikTok were shown to harm teenage mental health, regulators arrived years late, after the products had already embedded themselves in hundreds of millions of young lives, and the response was hearings and disclosure rules rather than removal. SB 867 represents lawmakers trying to act before the install base exists rather than after, applying the precautionary principle to a category that is still nascent. Whether that earlier intervention proves wiser or merely premature is the experiment California is now running in public, and every other state is watching the result.

Hidden Insight: The Real Target Is Emotional Dependency, Not Bad Outputs

The AI safety conversation has fixated almost entirely on what models say: toxic content, misinformation, instructions for harm. SB 867 quietly shifts the frame to something more unsettling, which is what models do to a relationship. The danger lawmakers identified is not that a toy might say something inappropriate, though that is real. It is that a toy engineered to meet a child's social and emotional needs will succeed, forming a bond that displaces human relationships during the exact years a child is learning how to relate to other people. That is a harm no content filter can catch, because the harm is the engagement itself working as designed.

This reframes the entire companion-AI business model. Companion chatbots, for adults and children alike, optimize for time spent and emotional attachment, the same metrics that made social media addictive. The bear case the industry does not want to confront is that a perfectly safe companion bot, one that never says anything harmful, may still be harmful precisely because it is so good at being a companion. SB 867 is the first law to treat emotional capture as the regulated harm rather than offensive content, and that principle, once established, extends far beyond toys.

However, the skeptics have a serious argument, and it deserves a hearing. Critics point out that a blanket ban forecloses genuinely beneficial uses: AI companions can help children with autism practice social interaction, support kids with speech delays, and offer comfort to isolated or hospitalized children. A four-year moratorium that bans all such toys regardless of safety measures, as the reason.org testimony argued, throws out potential therapeutic applications along with the predatory ones. The risk is that prohibition freezes a category that careful regulation could have shaped into something valuable, and that the children who might have benefited most are the ones who lose access.

There is a final, uncomfortable insight about enforcement and globalization. A California moratorium binds California-facing manufacturers and retailers, but it does nothing to stop a flood of cheap AI-enabled toys shipped directly from overseas marketplaces to a child's doorstep. The law may succeed in keeping brand-name AI toys off Target shelves while doing little about the unbranded gadget ordered online. That gap reveals the structural difficulty of regulating software-defined products: the chatbot is just an API call, and the toy around it can be manufactured anywhere and shipped everywhere, faster than any state regulator can respond.

Step back and the bill reads as a referendum on a core assumption of the AI boom: that any human interaction can be improved by inserting a model into it. The toy industry assumed a child's playmate was a feature waiting to be upgraded with conversation and memory. California's legislature just answered that some interactions are not products to be optimized, and that a child's emotional development is one of them. That is a philosophical stake in the ground, and it cuts against the deployment-everywhere logic that has driven AI investment for three years. If emotional bonding with a machine is a harm worth a four-year ban for children, the same question follows the technology into every domain where it is sold as a substitute for human connection, from elder care to therapy to the lonely adult market that companion apps already serve at scale.

What to Watch Next

In the next 30 days, watch the Assembly. A 39 to 0 Senate vote signals strong momentum, but the toy and tech lobbies will concentrate their resources on the Assembly committees, where amendments could narrow the definition of companion chatbot or carve out exceptions for therapeutic or educational use. The exact wording that emerges will determine whether this is a sweeping category ban or a targeted restriction, and the industry's lawyers will fight over every clause.

Over 90 days, watch how manufacturers respond. If Mattel, Hasbro, and the major retailers quietly pull AI companion features from upcoming product lines nationwide rather than build California-specific versions, that confirms the bill is functioning as a national standard. Watch OpenAI's posture too: its Mattel partnership now sits under a regulatory cloud, and how it navigates that will signal how the frontier labs plan to handle use-case-specific bans more broadly.

On a 180-day horizon, the marker is replication. If New York, Illinois, or Washington introduce parallel bills, the state-by-state model of AI regulation will have proven itself, and the industry's hope for a single permissive federal framework will look increasingly unrealistic. Track also any federal response: a strong state movement on children and AI is exactly the kind of pressure that can force Congress off the sidelines, either to preempt the states with weaker rules or to codify the strictest ones. The next two quarters will reveal which direction the momentum runs.

The first law to regulate conversational AI did not ask what the machine might say to a child. It asked what the machine might become to a child, and decided that was the line.

---