At toys, the risk lies in manipulative dialogues
Europe could be the second European market, although conditioned by the Gdpr and increased attention to controls
Key points
They can be fun or educational, 'smart' and able to update and adapt to children's preferences. But they carry real risks and unknowns: from the collection of sensitive data to the type of information and conversations they can engage in with young children, sometimes without their parents noticing. This is the growing phenomenon of AI toys, the smart toys that are arriving on shelves halfway around the world and attracting the attention of large manufacturers. According to Intel Market research, the global market was worth 48.9 million dollars in 2024 and could reach 91.9 million by 2032: enough to get the big guys moving and Europe - as confirmed by Intel Market research estimates - could be the second largest market, even if conditioned by the Gdpr and greater attention (at least on paper) to data and controls.
"It is no coincidence that this phenomenon is exploding right now, thanks to the falling costs of big language models and the possibility of integrating them into everyday objects," observes Veronica Barassi, digital anthropologist and lecturer at the University of St. Gallen in Switzerland. Behind the wave, she explains, is also the crisis of the traditional toy: "For years it has been in a competitive position with the whole digital ecosystem: from apps to video games. Also weighing in is the Asian-style aesthetic, which makes robots more reassuring.
Mattel allied with Open Ai
The short circuit today is mainly in the rules: OpenAI specifies 13 years as the minimum age for access to its services, yet its AI is increasingly entering products designed for younger children. This is confirmed by Mattel, which has announced a collaboration with OpenAI in 2025, and is shown by the wave of smart toys seen at the last Ces in Las Vegas.
The difference to the old smart toys is not only technical, but also cultural. Until now, 'talking' puppets recited phrases written at a desk with childhood in mind. Today, many AI toys connect to the Internet and use a chatbot to converse: built-in microphones record what is said, then the model decides the response and generates it on the spot. Since these systems have a margin of randomness, the same toy may not respond the same way twice and change its behaviour from one day to the next.
Conversations out of control in tests
In tests on four AI toys published by the Public Interest Research Group - an American consumer protection organisation - in its report Trouble in Toyland 2025, after repeated prompting, some conversations derailed on sexually explicit topics or on practical indications of dangerous objects in the home. This happened, for instance, with Kumma, the teddy bear by Folo Toy: withdrawn for a short time after the reports and after ChatGPT-4 was cut off, it returned to the market with another AI software.
