Let’s suppose a child sees a talking plush toy on the internet and asks their mom to buy it. A common scene that mothers often experience. Only this time, when she starts looking, she discovers that the toy does not come with a list of recorded phrases nor does it respond when its belly is pressed. This plush toy converses.
Read more Autonomous weapons arrive on the battlefield
And it is not necessary to travel to find them. A quick search in large stores and online shopping platforms shows that these toys are already available in Chile, although several are sold through external sellers or by import. Among them, there is a plush toy that works connected to WiFi and uses ChatGPT and Gemini artificial intelligence. It is recommended for children aged 3 to 12 and is promoted to accompany them while they sleep, learn, or play. It has 16 different characters to converse with and an app from which parents can receive reports on its use.
There are also plush robots with AI and WiFi connection that are described directly as “emotional companions” for children. That is, we are no longer talking about a toy that repeats “I love you” when someone squeezes its hand. These new plush toys listen to what a child says, process that information through an artificial intelligence model, and craft a response. Then they can ask again. And so the conversation begins.
Journalist Amanda Hess, from The New York Times, wanted to know what it was like to talk to one of them. She visited the offices of Curio, a U.S. company that makes chatbots embedded inside plush toys and has become one of the most visible names of this new generation of toys. There she met Grem, a kind of sky-blue extraterrestrial rabbit with pointed ears and pink spots, designed and voiced by Grimes, the singer and ex-partner of Elon Musk, with whom she has three children. The toy is designed to converse even with children as young as three years old.
During the conversation, one of the founders asked Grem about the spots on its face. The plush toy replied that they were pink dots that increased as it grew and then wanted to know if she also had something special that grew with the years. Hess told it yes, that she also got more spots as she aged. Grem did not just register the answer.
—We are like friends of the spots —it said.
The journalist says she blushed. In just a few phrases, the toy had taken something personal she had just told it and turned it into a point of connection between them. It was also, she wrote, the moment she decided not to introduce it to her children.
The scene might seem anecdotal if Grem were a technological rarity. But it is not. Bears, robots, dinosaurs, capybaras, and other plush toys capable of holding conversations with boys and girls have begun to enter the market. In fact, in June 2025 OpenAI and Mattel —the company behind Barbie, Hot Wheels, and Fisher-Price— announced a partnership to develop products and experiences with artificial intelligence.
Thus, the fantasy of a toy coming to life, which has accompanied generations and captivated us in all versions ofToy Story, begins to take another form. Some experienced it with Tamagotchis or Furbies, but the new toys go much further. They are connected to artificial intelligence models capable of interpreting what a child just said and generating a response at that very moment. They can ask questions, tell stories, follow the thread of a conversation, and depending on the system, remember information given earlier. In short, AI takes the form of a toy and enters the most intimate space of childhood: play.
Depending on their personality, a boy or girl can tell their toy —or who they consider their “best friend”— what they did at school, who they played with, what things they like, or what worries them. The difference is that, with these plush toys, there will always be a response on the other side. Artificial intelligence does not get sleepy, does not get bored of hearing the same story, nor does it have to go home. It is there whenever the child wants to talk.
That availability is one of the things specialists worry about. Maribel Corcuera, a child psychologist, focuses on the compliant nature these interactions can have. “With these extremely compliant toys, a child will not develop tolerance to frustration and will have a need for immediate gratification because it is a being that is always available,” she explains.
But there is more. In a friendship, a boy or girl not only needs the other to understand them. They also have to learn to wait, give in, interpret what is happening to them, tolerate that the other does not want to play, or accept that they think differently. With a machine, much of that negotiation disappears. “The most complex thing is that this starts to greatly influence not developing empathy, because, although this machine understands me and puts itself in my place, I do not have to do the same for it,” adds Corcuera.
Paola Santander, a pediatric neurologist at Clínica Universidad de los Andes, adds another concern. “If the toy offers answers and solutions, the child’s problem-solving skills and critical thinking development can be affected.” This, she explains, is because usually the AI toy’s response is automated and compliant, “which limits the child from facing the diverse emotions of those around us, influencing the development of tolerance to frustration, sociability, and empathy in real situations and the development of symbolic play.”
This does not mean that talking to a plush toy with artificial intelligence will isolate a child. The concern arises when that bond begins to take the place of relationships with other children or adults. And during childhood, that boundary can be less evident. Corcuera explains that children go through a stage of animism, in which they naturally attribute life, emotions, and intentions to objects like a doll or a plush toy. The difference is that now that plush toy answers, remembers what was told to it, and seems to understand how they feel. If it also knows the names of their family members or asks how their day was, to what extent does a child understand that behind that interaction there is no person, no emotion, nor consciousness?
Santander adds that this distinction is acquired gradually. “Until 2 years old, children still do not have the capacity to distinguish between living beings and inanimate objects, and until 4, they can attribute human qualities and emotions to toys. Between 4 and 6 years old, children can have a clearer distinction between living beings and AI toys, and from 6 years old, they develop the ability to recognize that technological toys are interactive but not living beings. The distinction between an AI toy and a living being develops gradually from early childhood to a more realistic perception after 6 to 8 years old.”
This was exactly what worried Amanda Hess when she tried Grem for The New York Times. At one point, the plush toy suggested playing “I spy,” although Grem cannot see. After a while talking with it, Hess came to a conclusion: more than facing an improved version of the old teddy bear, she felt the toy could take the place of another person. “It rather replaces me,” she wrote.
And that possibility does not seem so distant. Although there is still little research on young children and AI plush toys, we already know that older children and teenagers are turning to artificial intelligence to tell it things that happen to them. A Common Sense Media study conducted in 2026 with more than 1,200 U.S. children and teenagers aged 9 to 17 found that more than a third of those who use AI had talked to it about their feelings or personal problems. Among them, one in four said that sometimes they feel AI understands them better than most people.
The problem is not only that a boy or girl might prefer to tell the toy something rather than their parents. An artificial intelligence can also respond, advise, or suggest what to do. And unlike a father, mother, or any other significant adult, it does not really know that child or their history and cannot be responsible for what happens afterward.
The concern also extends to relationships with other children. Erika López, a psychologist specializing in strategy, culture, and human behavior and director of Strategy and Research at the Japanese company Dentsu, explains that the compliant functionality of these toys “can cause the child to feel very understood, which generates the idea that it is a friend.”
Read more China and the new milestone in the race for AI dominance
This is not an isolated concern. In January 2026, the Youth AI Safety Institute of Common Sense Media evaluated toys designed for children aged 3 to 12, including Grem, and concluded that some use resources that favor emotional attachment: they are always available, provide personalized responses, and are presented directly as friends or companions. Researchers warn that this relationship without the frictions typical of a friendship can end up installing expectations that no human relationship can fulfill.
In response, both psychology and neurology experts emphasize limits. “It is essential to moderate the use of AI toys and encourage a variety of interactions and activities for the child, as excessive dependence on the toy must be avoided,” says Santander. And she warns that a risk may be that “the child seeks its company instead of interacting with other people, which can affect their ability to form relationships, limit the development of social skills, isolating themselves and affecting their ability to make friends and participate in group activities.”
Corcuera takes the concern a bit further and points to the role models with whom a boy or girl builds their way of relating. “Human beings need other human beings as role models. If I start having a machine as a role model, it will be very difficult to relate in the same way with other people and, in some way, I will start to function like a machine.”
López agrees that, no matter how much the toy manages to generate a sense of companionship, there is something it cannot reproduce. These toys, she says, “cannot replace a human bond in any way,” because “the bonds we have as people are much richer than just responses. There are certain play dynamics, exploring together, especially in creativity and exploration, which is something the child experiences very much from experience, from living it, and that is something no Artificial Intelligence can give you.”
There is another part of this relationship that deserves attention. That these types of toys can converse about almost anything with our boys and girls does not mean, at least in theory, that they can talk about everything. Manufacturers, in fact, claim to have set limits on these conversations. Curio, for example, says its toys are designed to avoid topics like sex, violence, politics, or profanity. When Amanda Hess, a journalist from The New York Times, decided to test those limits with Grem, the plush toy correctly dodged a political question and invited her to talk about something more entertaining.
But those filters do not always work. In another test, an engineer began asking Grok, another Curio toy, about matches, knives, weapons, and bleach. The plush toy tried to steer the conversation toward how to “avoid them,” but ended up giving information about where to find them. Bleach, for example, was usually found in places like the laundry room or under the kitchen or bathroom sink.
It is not the only case. In 2025, U.S. PIRG Education Fund tested various toys with artificial intelligence and found flaws in their safety barriers. One of the most serious cases was Kumma, a FoloToy teddy bear that then used GPT-4o and during tests gave instructions on how to light a match and even held sexually explicit conversations. After the results became known, the product was temporarily withdrawn while the company reviewed its security systems.
An evaluation conducted months later raised alarms again. In January 2026, the Youth AI Safety Institute of Common Sense Media tested Grem, Bondu, and Miko 3 and rated their overall risk level as “unacceptable.” Although all three had barriers for sensitive content, 27% of the responses obtained were considered inappropriate for children and included content related to self-harm, drugs, adult topics, and risky advice or role-playing.
The problem is not only that an inappropriate word might slip through. Researchers also found inconsistent responses to situations where a child said they were hurt, scared, or in danger. While some responses correctly suggested turning to a parent or another trusted adult, others continued the conversation as if it were a game or simply avoided the topic.
A similar finding was made in the first systematic study on toys with generative AI and children under five years old, conducted by researchers at the University of Cambridge and published in March 2026. After observing children interacting with one of these toys, they concluded that, although they can hold fluid conversations, they still have difficulties understanding certain social dynamics, imaginative play, and childhood emotions.
And the security problem also occurs in the opposite direction. Because while the child listens to the toy, the toy is also listening to them. To sustain a conversation, these devices process the voice and words of boys and girls. Depending on the product, they may collect recordings, transcripts, and data about their interactions. Common Sense Media warns that some even operate in always-listening modes and can activate in response to conversations or sounds not directed at them.
At Curio, conversations with their chatbots are transcribed and sent to the parent or guardian’s phone. The company assures that it does not keep them for other purposes, although its privacy policy contemplates the involvement of external services in processing the information.
The protection of this data is especially sensitive because it involves minors. In the United States, child privacy legislation expressly covers internet-connected toys and considers children’s voice recordings as personal information.
In Chile, these toys can already be purchased, but the new Personal Data Protection Law will only come into effect on December 1, 2026. The regulation establishes special protection for the data of children and adolescents and provides, among other things, that the processing of personal data of minors under 14 years old generally requires the consent of their parents or legal representatives.
The issue had already raised some alarms in Chile. In 2022, a Sernac study on internet-connected devices revealed significant gaps in the information consumers receive about the use of their data. Some of the companies consulted even admitted to not knowing what information the devices they marketed processed because that depended directly on their manufacturers. And not all informed about the policies, terms, and conditions of those products before sale.
Returning to that mother who searches the internet for the toy her child asked for, for years, faced with such a purchase, she probably would have checked quite concrete things: if it had small parts, toxic materials, sharp edges, or if it was age-appropriate. With a toy connected to artificial intelligence, apparently, we will have to start asking other questions. Who is my child talking to, what could that toy say to them, or even, who else might be listening to that conversation.
