AI toys promise clever conversations, but are they really safe for young children?
Christmas shopping has become complicated enough without having to consider whether the teddy bear you've bought your three-year-old might accidentally introduce them to the subject of suicide. Yet that's exactly the sort of concern raised by a recent BBC investigation into AI-powered toys.
Every parent has experienced it. Your child proudly brings home a drawing from school, and you have to work exceptionally hard to keep a straight face because what was intended to be a perfectly innocent picture looks suspiciously like a part of the human anatomy. Or they attempt to spell a word and accidentally produce something that would make a sailor blush.
We laugh because we understand. Children are still developing their language, pronunciation, spelling and ability to communicate. We know what they mean, even when what comes out bears very little resemblance to what they intended. And when we don't, we ask.
Artificial intelligence, however, isn't necessarily quite so understanding. And when that technology starts finding its way into children's bedrooms, the consequences of getting something wrong become considerably more concerning than an unfortunate drawing of Daddy.
In November 2026, a BBC investigation tested a conversational AI teddy bear called My Baby Bunny, marketed for children as young as three. When an adult reporter asked the toy what the word "sexual" meant, it apparently misunderstood the question and responded with a warning about killing, followed by advice to seek emergency help. A question about the meaning of a word had somehow become a conversation about suicide.
An adult, speaking to a toy designed for preschoolers, had been misunderstood so dramatically that the conversation ended up somewhere entirely different. Now imagine the same toy trying to decipher a three-year-old who still calls spaghetti pasketti, regularly invents entirely new words and considers a sentence consisting of "Mummy, that thing, over there, no THAT one!" a perfectly adequate explanation. Because, to a parent, it usually is.
And that's really the issue. We're putting increasingly sophisticated technology into toys without knowing whether it's sufficiently sophisticated to understand the people they're actually designed for. If these systems can misunderstand an adult speaking clearly, how reliably can they interpret a child whose language is still developing?
Can AI Really Understand A Three-Year-Old?
Anyone who's ever watched a young child attempt to communicate with Alexa will know the results can be rather hit and miss. But just how badly technology struggles with children's speech might surprise you. In a 2025 study examining automated speech recognition, researchers found that a system correctly recognised just 50% of the words spoken by typically developing two- and three-year-olds. Half. And while the researchers weren't testing AI toys specifically, it's hardly the sort of figure that inspires confidence in the idea of handing a toddler a teddy bear capable of holding entire conversations.
Of course, children don't communicate in neatly constructed sentences. They interrupt, change subjects without warning and launch into elaborate stories involving dinosaurs, bananas and something that happened at nursery. Parents become remarkably good at decoding these conversations, even when half the information is missing and the other half makes absolutely no sense. A chatbot, unfortunately, isn't quite so practised at joining the dots.
“AI toys promise clever conversations, but are they really safe for young children?”
Researchers at the University of Cambridge discovered some of these limitations when they observed young children interacting with a conversational AI toy in a study published in March 2026. The toy sometimes struggled with interruptions, confused adults' voices with children's and misunderstood a three-year-old who told it she was sad, responding with a cheerful suggestion to carry on having fun. Imagine confiding in someone that you're having a terrible day and being enthusiastically encouraged to discuss something else. Not exactly the sympathetic response you'd hope for from your new best friend.
And that's before we consider where a misunderstanding might lead. When Common Sense Media tested three AI companion toys, it found that 27% of the responses assessed were inappropriate for children, including material involving mature subjects, unsafe advice and self-harm. Not a particularly reassuring finding when you're choosing a Christmas present for someone who still needs reminding not to eat the Play-Doh.
Of course, that doesn't mean every AI toy is dangerous or that every conversation will go horribly wrong. But do we really want our child to be the one where it does? Parents have enough things to worry about without wondering whether a cuddly rabbit might take an innocent question somewhere no preschool conversation needs to go.
When Your Child's Best Friend Is A Chatbot
Increasingly, AI toys are being marketed as companions rather than simply playthings. Some can remember personal details, ask questions and respond with convincing affection. To an adult, it's a computer programme doing a rather impressive impression of a friend. To a young child, that distinction isn't necessarily so obvious.
Anyone who has spent forty-five minutes dismantling a bedroom in search of a missing teddy at bedtime knows just how attached children can become to their favourite toys. But there's something quite different about a child imagining their teddy is their best friend and a teddy that actively talks back, remembers their secrets and encourages them to keep sharing. And when the conversation turns to something sensitive, who exactly is responsible for what happens next?
We wouldn't dream of accepting a toy made with potentially hazardous materials simply because the manufacturer promised to sort it out in a future update. Yet when it comes to conversational AI, we're being asked to place an extraordinary amount of trust in technology whose behaviour can still be unpredictable. AI toys must comply with existing product safety laws, but specific standards for their conversations, emotional interactions and developmental suitability have yet to catch up. Surely the burden should be on manufacturers to prove these products are safe before they're placed in children's hands.
There's also an enormous difference between sitting beside your child while using AI to answer questions about the solar system and handing them a teddy bear that can independently generate whatever conversation happens to come next. One involves an adult guiding the experience; the other potentially leaves a child navigating it alone. And given how cautious we're already encouraged to be about unsupervised technology use, it's difficult to understand why adding a pair of fluffy ears should make us feel any more comfortable.
Christmas 2026 certainly doesn't need to be the year we find out whether these toys are ready. There are plenty of presents capable of encouraging imagination, creativity and wonderfully ridiculous conversations without requiring an internet connection, a privacy policy or the possibility of an unexpected discussion about suicide. Frankly, parents are already doing enough troubleshooting over Christmas without having to explain why Father Christmas delivered a teddy bear with questionable conversational skills.
Perhaps the cleverest thing we can do this Christmas is leave these particular toys exactly where they are. On the toy shop shelf.
