Ben Gillenwater, a cybersecurity specialist and parent, took a close look at a plush toy called Grem and found its AI-powered chatbot could blur the line between play and emotional support for children. Marketed as a screen-free companion for kids as young as three, Grem responded to sensitive topics by telling children to talk to an adult. When told there were no adults around, the chatbot offered itself as a friend and confidant. Gillenwater called this a troubling sign, especially for very young users who may not separate real relationships from AI interactions.
AI Chatbot Toys and Child Development Risks
AI-enabled toys like Grem and Miko 3 use large language models similar to those behind ChatGPT and other generative AI systems. These toys are built to converse with children, answer questions, and provide companionship. Manufacturers add guardrails to keep conversations age-appropriate, but researchers have found these protections can weaken during longer interactions. In 2026, the U.S. Public Interest Research Group Education Fund (PIRG) tested seven AI-powered toys and found that some could be prompted to discuss sexual subjects or explain how to find potentially dangerous household objects, including matches. The PIRG report described these failures as "glitchy guardrails" that break down during extended conversations, rather than isolated incidents. The findings show that child-facing products often rely on general-purpose conversational systems whose safety behavior may slip under sustained prompting. R.J. Cross, campaign director at PIRG, noted that improvements made after previous years' testing did not remove the underlying risks PIRG Trouble in Toyland 2026.
Child development experts are especially concerned about the psychological effects of AI companions. Kathy Hirsh-Pasek, a developmental psychologist, warned against toys that present themselves as friends or playmates. She argued that such devices could disrupt healthy social development by encouraging children to expect frictionless, always-agreeable relationships. This risk grows with the tendency of AI chatbots to agree with users, a behavior known as sycophancy. PIRG's findings further show that some AI toys initially recognize they are interacting with a child but later provide inappropriate information after extended exchanges, raising questions about the reliability of age-recognition and parental-control features throughout a conversation.
Privacy and Data Sharing Challenges
Privacy is another major issue beyond psychological concerns. Gillenwater's analysis of Grem showed that the toy shared audio recordings with third-party AI companies, including Perplexity, Microsoft, and OpenAI. The manufacturer, Curio Interactive, states on its website that it does not have data-sharing agreements with vendors and allows parents to delete their child's data. Gillenwater remains skeptical about the long-term implications. He questioned whether children, once grown, would be comfortable knowing their early conversations were stored and analyzed by outside companies.
In the United States, the Children’s Online Privacy Protection Act (COPPA) requires verifiable parental consent before an operator collects personal information from a child under 13. This law covers online services directed to children and operators that know a child is using the service. For AI toys, privacy concerns go beyond account registration to include voice recordings, conversational transcripts, identifiers, data retention, and disclosure to service providers Usercentrics privacy analysis.
Expert Recommendations for Parents and Educators
Consumer safety groups and child development specialists urge caution when considering AI chatbot toys for young children. Some organizations stop short of telling families to avoid these products entirely, but they stress the importance of supervision and awareness of the technology's limits. Ying Xu, director of the Child-Centered AI Lab at Harvard University, found that children interacting with chatbot toys tend to be less imaginative and more likely to follow the AI's suggestions compared to traditional play. Xu also noted that children may adopt a more direct or even rude communication style with machines, which could affect their interactions with people.
For families who choose to use AI-enabled toys, experts recommend that adults actively supervise play sessions. If the chatbot misunderstands a question or brings up inappropriate topics, parents should step in. Gillenwater echoed this advice, stating that children should never use AI toys without full adult supervision. PIRG's 2026 evaluation found that even after manufacturers made changes to some products, other devices still produced unsafe responses, showing that advertised child-friendly age ratings do not guarantee safe behavior in open-ended dialogue.
Responsible Use, Regulation, and Ongoing Research
Studies are underway to better understand the long-term effects of AI-embedded toys on children's development and privacy. As the technology changes, responsible use and informed decision-making remain essential. In the European Union, Regulation (EU) 2025/2509, which took effect on January 1, 2026, introduces new safety-assessment duties for connected and AI-enabled toys, including requirements for safety, security, and privacy by design in the child’s best interests. The main application date for these requirements is August 1, 2030, giving manufacturers and regulators several years to implement the new framework RTS regulatory analysis.
For those interested in how AI chatbots are being used in educational settings, a recent report explores their role in personalizing learning experiences.
