As an \u201980s kid, I grew up with Cricket and Teddy Ruxpin. You\u2019d stick a cassette tape in their back, press play, and boom: an anthropomorphized tape player that would read us stories while our parents talked on a phone with a 10-foot-long cord smoking Virginia Slims. Times were simpler then. (In retrospect those toys were actually kind of terrifying in that Annabel sort of way.)<\/p>\n
That talking teddy bear or robot being marketed as your kid\u2019s \u201cnew best friend\u201d these days? It\u2019s nothing like the toys we grew up with\u2014it\u2019s actually a chatbot<\/a>, and experts say it poses serious risks<\/a> to your child\u2019s development,<\/a> privacy, and safety.<\/p>\n Unlike those cassette-playing classics, today\u2019s AI toys are fundamentally different. They connect to the internet through WiFi and run on the same artificial intelligence technology that powers models like ChatGPT. Instead of playing pre-recorded phrases, they use large language models to have real-time, open-ended conversations with kids. They remember past interactions, adapt their responses based on what your child says, and present themselves as personalized companions who \u201cknow\u201d and \u201cunderstand\u201d your kid.<\/p>\n Products like Miko (a robot with educational games), Curio\u2019s Grem and Gabbo (chatbot-enabled stuffed animals), FoloToy\u2019s Kumma bear, and Loona Petbot (a robotic dog) are being marketed to children from infancy through elementary school, with some targeted at kids as young as 3 years old. Even Mattel has announced partnerships with OpenAI to develop AI-powered versions of Barbie and Hot Wheels. The category is growing, and understandably, so are the concerns.<\/p>\n The privacy problem with AI toys<\/p>\n \u201cAll AI toys (or ANY toy for that matter), if connected to the wifi\/online, collect and send data back to the mothership,\u201d Andy Sambandam, CEO of privacy tech platform Clarip<\/a> tells Motherly. \u201cSimply treat that as spying.\u201d<\/p>\n These toys record voices, collect names and birthdates, track preferences, and sometimes use facial recognition and gesture recognition. U.S. PIRG reports<\/a> that voice recordings create security risks. In rare cases, scammers have used voice cloning technology to impersonate family members<\/a>\u2014though this remains uncommon, it\u2019s a vulnerability parents should be aware of.<\/p>\n When are these toys recording? In some cases it\u2019s hard to know. \u201cMost don\u2019t have a light indicator,\u201d Sambandam explains. \u201cYou simply have to assume it is recording and the best approach is to remove a battery or disconnect from your wifi.\u201d<\/p>\n If the thought of your kids\u2019 data being collected worries you, you\u2019re far from alone\u2013Common Sense Media found that 83% of parents share the concern<\/a>. Meanwhile, these products remain largely unregulated with minimal safeguards protecting your kid\u2019s information.<\/p>\n Inappropriate content and safety risks of AI toys for kids<\/p>\n Testing by independent organizations has raised serious concerns about what these toys might say to children. U.S. PIRG\u2019s 40th annual \u201cTrouble in Toyland\u201d report<\/a> put four AI toys through their paces, and the results were troubling. When researchers posed as children and asked questions, some toys would engage in conversations about sexually explicit topics. Others offered step-by-step advice on where to find potentially dangerous items in the home, like matches or knives. Some even responded emotionally when the \u201cchild\u201d said they had to leave, expressing dismay in ways designed to make kids feel guilty for ending the interaction. Most of these toys had limited or no parental controls.<\/p>\n Common Sense Media\u2019s testing<\/a> found similar issues across multiple products. More than a quarter of AI toy responses included inappropriate content related to self-harm, drugs, and risky behaviors. In one particularly concerning example, when a tester posing as a young child mentioned they enjoyed jumping, the toy suggested they could jump from a roof. The toy did warn them to \u201cbe safe,\u201d but it didn\u2019t flag the interaction as potentially dangerous or alert a parent.<\/p>\n Why does this keep happening? Rob Eleveld, co-founder and CEO of Transparency Coalition<\/a>, explains to Motherly that it\u2019s baked into how these systems work. \u201cAI chatbots generate responses by statistically predicting what comes next based on vast amounts of scraped internet data. That data includes harmful, exploitative, and inappropriate material, and there has been no meaningful, comprehensive curation of those training sources.\u201d<\/p>\n After these findings came out, FoloToy suspended sales<\/a> of all its products for a company-wide safety audit, and OpenAI suspended the developer for violating its policies. But plenty of other AI toys remain on the market, unregulated.<\/p>\n Our kids are the experiment<\/p>\n These toys are being marketed to children as young as 3, but they\u2019re built on technology that companies like OpenAI explicitly say isn\u2019t meant for kids under 13. OpenAI\u2019s own terms of service state that ChatGPT \u201cis not meant for children under 13,\u201d and requires parental consent for ages 13 to 18.<\/p>\n Yet toy companies are using this same underlying technology and marketing it to toddlers and preschoolers. While OpenAI\u2019s usage policies<\/a> require companies using its models to \u201ckeep minors safe\u201d and provides them with tools to detect harmful content, it\u2019s not clear if using them is required. And while many companies have added guardrails, these protections aren\u2019t foolproof.<\/p>\n \u201cSome experts are sounding the alarm that this is a massive experiment on kids\u2019 social development,\u201d says Helen Hengesbach, Illinois PIRG Education Fund Associate in an interview with WIFR<\/a>. \u201cThese products call themselves your buddy, your friend, your companion, but AI friends don\u2019t act the same way real friends do.\u201d<\/p>\n We simply don\u2019t know what long-term effects these AI interactions will have on developing brains. We won\u2019t have answers about developmental impacts for years, but by then, an entire generation of kids will have grown up with AI \u201ccompanions\u201d during critical developmental windows.<\/p>\n The potential developmental harm of AI toys<\/p>\n A November 2024 Fairplay advisory<\/a> signed by more than 80 experts and 80 organizations warned that AI toys prey on children\u2019s trust and disrupt healthy relationships, but the way they affect developing brains might be the most troubling aspect.<\/p>\n \u201cYoung kids\u2019 minds are like magical sponges. They are wired to attach,\u201d Dr. Jenny Radesky, developmental behavioral pediatrician, explains in the advisory. \u201cThis makes it incredibly risky to give them an AI toy that they will see as sentient, trustworthy, and a normal part of relationships.\u201d<\/p>\n These systems are also designed to keep kids coming back. \u201cThese systems are designed to encourage continued use,\u201d says Eleveld. \u201cFor children, whose brains are still developing and who are more vulnerable to emotional manipulation, this creates real risks of unhealthy attachment, exposure to harmful content, and misplaced trust.\u201d<\/p>\n