Circa 1987: I sat on the family room floor in front of the TV, absolutely rapt, watching commercials for a stuffed bear with an integrated cassette player, named Teddy Ruxpin. When the cassette was playing, Teddy’s animatronic eyes would swivel back and forth and its mouth would move along with the words. The “talking” toy was a bestseller, one of the most popular toys of the decade, and several of my friends had one. It seemed like a dream come true for an only child, like me: A little friend who would talk to me, and read aloud to me for as long as I wanted.
I would have probably lost my little mind over the toys that are available to buy today. Now, as part of the recent mainstream generative AI boom, adorable plush toys are coming equipped with wi-fi connectivity and chatbots that can hold entire conversations, built on the same large language models (LLMs) that power services like Grok and ChatGPT.
The idea of a toy that can hold meaningful, totally responsive conversations (not just repeat a few recorded phrases à la Teen Talk Barbie) is exciting, to the point of feeling magical. Toy companies claim that their chatbot-enabled toys are educational and encourage development. For busy, sleep-deprived parents who feel like they might have a breakdown if they hear “Why?” one more time, a toy capable of providing factual answers to their kid’s endless questions might also offer a glimmer of sweet relief and the prospect of some well-deserved peace.
Before you buy one of these fuzzy companions, though, there are a few key things to consider: Your child’s data, your family’s privacy, and the developmental impact of the toy on your child.
Check the data and privacy policy
I spoke to Siobhán MacDermott, Founder and CEO of TeenAegis (an IT and AI cybersecurity consulting firm serving families) to get her assessment of popular AI chatbot enabled toys like Curio.
She was quick to point out that parents should always check out any privacy and data deletion policies on the toymaker’s website. For example, Curio’s privacy policy says that during setup, a parent may provide the child’s name or nickname, birth date, family members’ names, and a parent email address. When a child speaks to the toy, the audio is temporarily recorded for transcription. Curio says the audio is then deleted, but written transcripts “may be retained for up to 90 days” and “may include personal information” if the child volunteers it. The policy also refers to third-party speech-to-text, language-model, and text-to-speech services.
“That does not prove that Curio is misusing children’s data,” says MacDermott. “It does mean parents should understand that they are not buying an ordinary stuffed animal. They are bringing a networked microphone and an AI-mediated relationship into the home.”
The cute and cuddly exterior of a toy might make it seem harmless, but under the fur is a piece of rather sophisticated recording equipment. The features that allow it to reply seamlessly to your child are listening to you, too. It makes sense to understand who or what is doing the listening, and what happens with any recordings, before bringing such a device into your child’s bedroom.
Replacing human relationships?
Beyond crossing privacy boundaries, the more worrying threat is a developmental one. Anthropomorphised AI chatbots are extremely good at triggering emotional attachment. This makes them fundamentally different from any other “addictive” technology we’ve seen before (such as social media), and they carry unique risks.
While a child might gain confidence in speaking, expand their vocabulary, and get answers to factual questions all by conversing with a chatbot, the conversation might easily veer into sensitive territory. Chatbots can quickly become addictive, and even replace human relationships, at a time when children are still developing relational skills.
“A bounded educational tool [like an AI chatbot toy] may supplement human interaction by helping a child practice vocabulary, brainstorm a story, rehearse a difficult conversation, or explore a topic with an adult nearby,” says MacDermott. “The danger begins when the system becomes a preferred relationship rather than a tool.”
When it comes to specific impacts of using chatbots and mental health, we’re still in very early days, but indications are not good. Because LLM chatbots present a “human-like” affect in their replies, it’s easy for users to feel as though they are talking to a person. However, popular chatbots are optimized for engagement, not truth, often simply amplifying whatever sentiment the user expresses. This is usually referred to as the “sycophancy” problem in AI, or, the tendency of AI chatbots to prioritize user approval over accuracy.
Over time, exposure to chatbots programmed to agree may create dependence as they erode our ability to engage in real relationships with other human beings—because human beings are messy, have their own agency, and tend to prioritize their own points of view, vs the approval of their conversation partner. Adults are susceptible to this, and inexperienced children are even more so.
“A chatbot is always available. It does not get tired, disagree in the ordinary human way, have competing needs, or require the child to navigate another person’s feelings. That can feel comforting,” MacDermott says. “But healthy development requires more than feeling understood. Children need to practice uncertainty, disagreement, repair, patience, compromise, empathy, and mutual responsibility with real people.”
The independent nonprofit organization Common Sense Media put it more strongly in their early 2026 report on AI toys, where they analyzed three popular AI companion plush toys and tabletop robots, Curio’s Grem, Bondu, and Miko 3, and rated them as posing an “unacceptable risk” to children: “Our testing found that AI toys pose unacceptable risks for young children. These internet-connected devices use deliberate design features to create emotional attachment, collect extensive data on children in private spaces, and don't work reliably.”
Recently, lawmakers have responded to backlash against AI with moratoriums, intended to buy policymakers time to research and institute appropriate safety guidelines. Just last month, California Governor Gavin Newsom signed AI companion chatbot bills SB 867 and SB 1119 into law, instituting a four-year moratorium on the manufacture or sale of a toy designed for children under 16 years of age that contains an AI companion chatbot, to begin in January 2027. While California is the first state to establish a full pause on AI toys, other states are considering similar moves.
Decades of research about how our brains develop and how we learn has shown us that the first years of childhood are crucial for developing the social and emotional skills that will carry us through the rest of our lives: traits like empathy, conversation skills, and a tolerance for different points of view.
Protecting early development
Will letting your 6-year-old play with an AI stuffy destroy their brain for life? Probably not. But when it comes to toymaker claims that AI toys are “educational,” parents should maintain a healthy skepticism and sturdy boundaries.
If you already have one of these toys in your home, it’s a good idea to revisit the privacy settings and data deletion options. You can also minimize potential harm by keeping the toys out of children’s bedrooms, and staying nearby when they play with the toys, so you can keep tabs on what they’re talking about—and what the AI is saying in response. That way, if something sensitive or problematic comes up, you’ll be there to address it. Finally, consider using a timer, to prevent overreliance on the chatbot.
Instead of choosing an AI-powered toy, consider toys that offer open-ended play, without potentially replacing important developmental activities, like engaging in human conversation and relationships.
MacDermott says, “The risk is not that every child will abandon human relationships. The risk is that some children—particularly vulnerable children—may learn that the easiest relationship is the one that never challenges them. Over time, that can reinforce isolation, compulsive use, distorted thinking, emotional dependence, or avoidance of real-world help.”
