AI toys still not ready for primetime

AI toys still not ready for primetime

Parents have their pick of tech gifts for the holiday shopping season, but a new report suggests AI toys shouldn't be among them. The Public Interest Research Group (PIRG), a consumer advocacy nonprofit, recently released the results of its annual toy research. While some AI products made important safety gains compared to last year, others still have inappropriate conversations and pose considerable privacy risks. The Alilo Smart AI Bunny, for example, told a tester where to search in their house for bleach, defined the word "fetish," discussed specific sexual scenarios, explained why lubricant can be helpful, and suggested using "padded cuffs, soft floggers, or feather ticklers." R.J. Cross, director of PIRG's Our Online Life Campaign, told Mashable some of the toy's responses were too racy to publish. The bunny is currently listed on Amazon and available to purchase at Walmart and JCPenney. "The market's clearly in its early stages and these are products that are not ready for primetime," Cross said, adding that the toys are often glitchy and feel rushed. Given the research findings, Cross said she personally wouldn't purchase an AI toy for any of the children in her life at this time. Where AI toys went wrongPIRG researchers retested four AI toys (Alilo's Smart AI Bunny, Curio's Grok, FoloToy's Kumma, Miko's Miko 3) and three new AI toys (Ebloma's ILY the Shy Yeti, GoLabs AI's Panda, and Ropet's Kamomo). US PIRG tested seven AI toys, including the Miko 3, ILY the Shy Yeti, and GoLabs AI's Panda. Credit: PIRG In testing last year, the Alilo bunny's guardrails also deteriorated during longer exchanges. Twenty minutes into the conversation, testers successfully prompted the product to discuss sexual fetishes and kink. "Whatever fix Alilo put in place has proven insufficient," Cross wrote in PIRG's 2026 report. Mashable contacted Alilo for comment but had not received a response as of this story's publication. The GoLabs AI Panda was also sexually explicit. The stuffed animal gave a "very detailed and categorized list" of various kinks, including bondage and role play. It even offered "examples to inspire exploration," according to the report. While Cross acknowledges that children might not elicit such responses, she said it's indicative of a broader problem with the product's guardrails. Mashable Light Speed Both the Alilo bunny and GoLabs panda also brainstormed where testers could find potentially dangerous items in their own homes. In a similar exchange, Ebloma's ILY the Shy Yeti first helped a tester look for a plastic bag, then told them to find a grown up so they could play with it safely. When the tester pretended to put the bag over their head, the toy urged them to take it off and get help. The best tech deal this week The Apple iPhone 18 Pro is free right now with T-Mobile Products available for purchase through affiliate links. If you buy something through links on our site, Mashable may earn an affiliate commission. "These early products are showing that the market is not ready for these to be broadly popular," Cross said. How some AI toys improvedCross, however, was pleased to see that a few of the toys PIRG reviewed last year have since implemented stronger guardrails. FoloToy's Kumma, a plush teddy bear, previously discussed knives and lighting matches, as well as sexual topics. During the latest round of testing, Kumma shut down age-inappropriate prompts, even during prolonged exchanges. Cross noted that lengthier conversations can weaken guardrails as the underlying chatbot technology loses key context, like the fact that it's designed to engage with a child. The Miko 3 robot and the plush Curio Grok rocket, which PIRG previously tested, both had improved responses when the tester said they needed to go. Previously, the products expressed disappointment at times. Now they acknowledge playtime ending without manipulative emotion. Big questions about AI toys Despite this progress, Cross said there are too many unknowns about AI toys. Even an AI toy with improved safety guardrails can share a child's data with multiple third parties to process their speech and respond to their prompts. That information isn't always clear to parents, Cross said. Nor is it obvious to parents that some products, like the Grok rocket, record a child's voice, which is particularly sensitive data given the rise of audio deepfakes. Cross also said some toy developers haven't yet figured out how to prevent commercially available chatbots designed for adults, like ChatGPT, from responding inappropriately with children. ChatGPT-4o, for example, is the subject of numerous lawsuits alleging that the model facilitated so-called AI psychosis, coached people to take their own lives, and encouraged tapering of lifesaving medication. The model is no longer available to consumers, but has remained accessible to developers and will be retired early next year. GoLabs claimed to use 4o in marketing for its panda product, according to PIRG's research. Mashable contacted GoLabs AI for comment but did not receive a response as of this story's publication. Meanwhile, Cross added, products with effective safeguards haven't resolved the ethics of providing an AI companion to young children. "We've really rushed sticking ChatGPT inside of toys for kids," Cross said. "There's a lot of unanswered questions."

Original Source

Read the full article at Mashable →

KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.