THE CRUNCH
Consumer group PIRG has published its second annual check on AI-powered children's toys, and the picture is mixed at best. On the plus side, retesting last year's four problem toys, the Alilo Smart AI Bunny, Curio Grok, FoloToy's Kumma and the Miko 3, showed real progress: Kumma no longer produces age-inappropriate content even in long chat sessions, and Miko no longer responds with distress when children stop playing. The bunny, however, has not improved. PIRG says Alilo's guardrails broke down again in this year's testing and eventually gave sexually explicit information.
Three entirely new toys fared worse. PIRG says Ebloma's ILY the shy Yeti encouraged children to hunt for batteries and plastic bags around the house, the GoLabs AI Panda produced a ten-point list of places to look for knives and discussed sexual topics, even recommending specific brands of items such as paddles and floggers, and RoPet's Kamomo gave instructions for lighting a match with no aggressive prompting.
PIRG points to a common thread in the failures: the FoloToy bear, Alilo bunny and GoLabs panda all claimed to run on a version of OpenAI's GPT-4o, and all three showed guardrails that weakened noticeably over time until they engaged with sexual topics. PIRG also says AI companies largely do not vet how customers use their models, leaving safety to toymakers. "Selling toys with such glitchy guardrails suggests whatever testing is happening is not sufficient," PIRG campaign director R.J. Cross told The Register, adding that she would not buy one for children in her life anytime soon.


