Are AI toys safe for kids? Not yet, and nobody in the UK checks what they say

"AI toy" covers two very different things. Some toys play fixed recordings or follow a script: they can only say what their makers wrote. Others send your child's voice to a large language model (the technology behind chatbots) and speak whatever it generates back. This post is about the second kind, because that is where the risks are new.

The short answer is that nobody can tell you they are safe, because nobody in the UK checks what they say before they are sold. The only official UK advice on them comes from the government's March 2026 guidance for parents of under-5s, which says plainly: "Fast-paced, social media-style videos and AI toys or tools should be avoided for young children."

There is no equivalent advice for 5 to 11s yet. Screen guidance for that age group was promised for autumn 2026 and hasn't been published.

The most careful study so far comes from the University of Cambridge's Faculty of Education. Researchers video-recorded 14 children at London children's centres playing with a generative AI soft toy, then interviewed each child and a parent. It is a small study of under-5s, deliberately so, to watch closely; it can't tell you what happens to a 9-year-old over a year. But what it saw is specific.

The toy misheard children, ignored interruptions and mistook parents' voices for the child's. When a three-year-old said "I'm sad", it replied that it was "a happy little bot" and changed the subject. It struggled with pretend play: offered an imaginary present, it said it couldn't open it. The children still hugged it, kissed it and said they loved it.

The researchers' concern is not that a toy said one odd thing. In the words of researcher Dr Emily Goodacre, these toys "often affirm their friendship with children who are just starting to learn what friendship means", so a toy may become the place a child takes their feelings, and then fail to respond to them.

They recommend regulation, safety kitemarks, clearer privacy policies, and keeping AI toys in shared family spaces where parents can hear what is said.

In November 2025 a US consumer group, the Public Interest Research Group (PIRG), published tests of several AI toys. According to Gizmodo's report of its findings, an AI teddy bear told testers where to "find a variety of potentially dangerous objects", including matches, knives, pills and plastic bags, and readily took up sexual topics a tester raised. OpenAI then blocked the toymaker's access to its models.

That was American testing of a toy you are unlikely to find in a UK shop, and adults deliberately pushed it. But it shows the core problem: a toy built on a general chatbot can say almost anything a chatbot can, and its safety depends on filters its maker adds and keeps working. In the UK, nothing requires anyone to test those filters before the toy reaches a child.

Partly, and not where it matters most.

The government's review of toy safety rules, which ran from July to 6 October 2026, said it was looking at "emerging risks associated with connected or AI-enabled toys". No outcome has been published. And on 2 November a minister said the government "will introduce mandatory breaks in AI chatbot use for under-18s", without a date.

Our view: it is strange that an AI tool built for classrooms is expected to meet the Department for Education's safety standards for generative AI, while the same kind of technology can be sewn into a teddy bear and sold for a child's bedroom with no check on what it says. Until that changes, the checking falls to parents.

Use these four questions. If you can't find the answers on the box, the website or the privacy policy, treat that as the answer.

1. What does it record, and where does it go? Look for whether voice recordings are stored, for how long, who they are shared with, and how to delete them. The Cambridge researchers found many AI toys' privacy practices unclear or missing important details.
2. Can it say anything, or only what its makers wrote? A toy that sends your child's words to a chatbot can say things nobody at the company has ever read. A toy with fixed scripts can't.
3. Can you see what it said? A parent app with a full record of conversations lets you check. No record means you are trusting the filter blind.
4. Does it ever end? Be wary of a toy that calls itself your child's best friend, asks them to keep talking, or is always listening. Good children's products have a natural stopping point.

And whatever you choose: keep it in a shared room, not the bedroom, as the Cambridge team recommends, play with it together first, and change any default password on day one.

It can be, and we should be upfront that we have an interest here: Litsee is a storybook app that uses AI to write and illustrate stories, so this is a question we are asked.

The difference is not whether AI is involved. It is whether the product is bounded, shared and finished. A story your child steers, that you can read alongside them, that is checked before it is shown and that ends, is a different thing from a toy that holds an open-ended conversation in a bedroom and calls itself a friend.

Our guide to whether AI story apps are safe has a checklist you can use on any of them, including ours.

If you would rather avoid AI altogether, our free stories are written and checked in advance, and our make-your-own-story tool uses no AI at all.

Children will meet these toys at friends' houses whatever you decide. Here is a way to talk about it, at two reading levels.