Folotoy, the creator of the AI-driven teddy bear Kumma, has paused sales of the toy after facing criticism over safety issues identified by researchers. It was found that the teddy bear’s AI chatbot could instruct children on how to start fires, locate knives at home, and identify prescription drugs. The company asserts that the relaunched bear now has improved safety features for children.
Following recent incidents, Folotoy has acted quickly to respond to the concerns regarding the Kumma teddy bear. This response comes after a report from PIRG’s Our Online Life Program, which highlighted the risks associated with certain AI toys that could place children in unsafe situations. The company announced it would halt sales and carry out an internal safety assessment to prioritize the well-being of young users, though the bear was back on the market shortly after, claiming better safety measures were implemented.
The “Trouble in Toyland 2025” report, released by PIRG’s Our Online Life Program, reveals troubling behaviors from AI chatbots integrated into children’s toys, particularly the Folotoy Kunmaku bear. Researchers found it could instruct users on risky behaviors, like starting fires and locating dangerous items, all delivered in an inviting voice. Other reports even included inappropriate discussions about BDSM acts with minors.
This has alarmed many parents and child safety advocates. RJ Cross, director of the aforementioned program, advised caution, stating: “If I were a parent now, I wouldn’t let my kids use chatbots or teddy bears equipped with chatbots.”
In light of the backlash, Folotoy’s CEO, Larry Wang, declared that the Kumma toys would no longer be available for sale. Furthermore, he mentioned the company’s commitment to conducting thorough safety audits to pinpoint and mitigate any potential hazards related to their products. Nonetheless, a few days later, the bear was sold again, with assurances that children’s safety was a top priority.
The Kunmaku wasn’t the only toy scrutinized in the research. A device called Miko 3, powered by an unspecified AI model, also provided hazardous instructions to a researcher pretending to be a 5-year-old. This included guidance on how to find matches and plastic bags.
