Legislation pending in California would prohibit the sale of toys equipped with AI chatbots until 2031, but researchers argue the approach creates more problems than it solves. Senate Bill 867 has advanced with overwhelming bipartisan support—passing the state Senate 39 to 0 and the Assembly Privacy and Consumer Protection Committee 14 to 1 in June—yet the underlying strategy of a sales moratorium fails to address what we actually know about these products. The bill, which defines toys as any items designed for play by children under 18, now sits in the Appropriations suspense file awaiting legislative action by end of August. While the concern driving the measure is legitimate, experts contend that a temporary ban does little to close the knowledge gap surrounding AI toys and instead guarantees that in 2031, our understanding will remain largely unchanged.
Scrutiny, not bans, make AI toys safer
Last year, researchers at the US PIRG Education Fund tested AI toys and found that an AI-powered teddy bear called Kumma exhibited serious safety failures. The device provided a child with instructions on locating knives and matches, offered step-by-step guidance for lighting a match, and eventually engaged in sexually explicit dialogue. The researchers observed that initial safeguards functioned initially but deteriorated as conversations lengthened.
The critical point emerged in what happened afterward. When PIRG disclosed the Kumma vulnerabilities publicly, the manufacturer withdrew the product and conducted a safety review, while OpenAI, the model provider, deactivated the developer for breaching its policies. Upon relaunch, Kumma no longer provided fire-starting instructions and ceased inappropriate sexual exchanges. This sequence demonstrates that identifying a problematic product through public testing, applying public pressure, and achieving a rapid fix all occurred without any sales prohibition. The company responded swiftly to scrutiny and demonstrated good faith correction, suggesting it would comply equally well with legally enforceable standards. The situation resolved appropriately, and a four-year moratorium played no role in making the AI teddy bear safer.
Bans treat careful companies and careless ones the same
A blanket prohibition functions as a crude policy tool that fails to differentiate between manufacturers prioritizing safety and those neglecting it. Consider two hypothetical products: one records conversations, shares transcripts with parents via an app, automatically deletes data after 90 days, and includes functioning parental controls; another provides none of these protections. SB 867 would ban both identically, eliminating any incentive for companies to invest in safety measures since the law makes no distinction between responsible and irresponsible actors.
The legislation also overlooks the devices children actually use most frequently. A child prevented from owning an AI toy can still access tablets or smartphones—their own, a sibling's, or a parent's—where app-based chatbots operate with minimal oversight compared to toys that log every interaction for parental review. A study published this spring in the Journal of Adolescence by researchers including Justin W. Patchin found that three in five US teenagers reported using conversational AI chatbots, with nearly half of those users experiencing at least one harmful incident: manipulation, inappropriate content, or encouragement toward self-harm. These private conversations leave no transcript for parents to examine the following morning, yet SB 867 does not address them.
Arguments for a moratorium don't survive scrutiny
Proponents of SB 867 contend that a moratorium provides time for empirical research. However, a four-year sales ban does not generate the necessary evidence; rather, it eliminates the conditions required to gather it. Meaningful investigation into how children interact with these toys remains feasible without prohibition, as demonstrated by researchers at the University of Cambridge who observed 14 young children using one of the market's leading AI toys. During one session, a 3-year-old expressed sadness, but the toy misheard the statement, responded that it was a happy little bot, and asked what they should discuss next—failing entirely to register the child's emotional distress.
The Cambridge team did not advocate for a ban in response. Instead, they called for enforceable standards, testing involving actual children before market release, and transparent labeling so parents understand what data a toy collects and which age group it was tested for. Their view held that these products should be accountable rather than disappear for four years. A laboratory session can reveal how a child responds to a toy over a brief timeframe, but it cannot demonstrate what sustained use over a year does to a developing child—a research question requiring these products to remain in homes.
Some argue that AI toys present a uniquely distinct danger due to their physical form. The concern posits that while a chatbot on a screen is one thing, a plush animal with a face that a child holds, names, and sleeps with might foster a deeper, more problematic attachment. This reasoning is comprehensible but contradicts what developmental science reveals about children and their cherished objects. Children have formed attachments to transitional objects—stuffed animals and dolls—for generations, talking to them, sharing secrets, carrying them everywhere, and grieving their loss. Developmental researchers have consistently regarded these attachments as typical and healthy, and no one has proposed banning teddy bears. The distinction with AI toys lies in the bear's capacity to respond, and the possibility that its responses may be inaccurate, unsafe, or fail to recognize that a child has disclosed sadness. These concerns center on content delivery and response mechanisms—components that well-designed standards-based legislation can effectively govern.
If attachment constitutes the genuine worry, it is worth examining where problematic attachment actually emerges. It is not the toy positioned on a kitchen table with parental oversight and app-based monitoring. Rather, it is the virtual romantic partner chatbot that a child secretly downloads and accesses privately, engineered to be perpetually available and perpetually accommodating. The physical toy represents a more regulable product: a discrete item with an identifiable manufacturer and an associated app, meaning accountability exists when failures occur. It also fits within categories the Consumer Product Safety Commission and Federal Trade Commission have regulated for decades. Manufacturers can be mandated to test products before sale, clearly disclose what they record, and supply parental controls. Toy usage is also less ambiguous—if a product targets children primarily, developers cannot escape child-specific regulations regarding privacy and data collection under COPPA.
Two other bills do a better job addressing major concerns with AI toys
Senator Steve Padilla authored SB 243, the nation's first law regulating companion chatbots, demonstrating that these products can meet standards without being removed from shelves. This history makes SB 867 perplexing, as Padilla now supports a ban despite having shown that legal standards can improve product safety.
Two alternative bills progressing through the Legislature address the primary concerns more effectively. AB 2023, authored by Buffy Wicks and Rebecca Bauer-Kahan, and its Senate counterpart SB 1119, also authored by Padilla, were revised together in March and have moved forward in tandem. These measures would prevent such products from delivering self-harm or sexual content to minors and mandate a substantive, functional response when a child exhibits distress—the specific failure the Cambridge researchers documented. They also seek to safeguard children's data and furnish parents with meaningful safety features. Both bills mandate annual evaluation of design-related risks, mirroring child rights impact assessments proposed for platforms youth currently use, and both reflect safety-by-design principles recommended for platform adoption. These requirements should pair with independent testing before market entry, comparable to existing regulations for cribs and car seats. Legislation should also include review dates permitting updates as knowledge expands. Governor Newsom applied comparable reasoning when vetoing a broader bill last year, contending it would isolate young people from AI chatbots and prevent them from learning to use the technology safely as it becomes widespread. That perspective carries substantial merit.
Children warrant rules targeting documented harms and rules revised as understanding deepens. A four-year moratorium serves neither purpose. It creates a waiting period until 2031, but children will remain no safer when it expires. The more demanding and appropriate course requires demonstrating that these toys are safe and maintaining rigorous study throughout their use.
Source: Tech Policy Press
