Thinking of buying an AI teddy this Christmas? Check these things first
Allen Clark

AI teddies are being sold as Christmas presents this year. They look like ordinary soft toys, but inside there's a microphone, a speaker and a connection to an AI chatbot. Your child talks to it and it talks back.
Last week the BBC tested two of them. Both are marketed at children from the age of three. One was Grem, made by the US firm Curio. The other was My Baby Bunny, also sold as Doubao, which you can buy online in the UK for £75. Grem handled awkward questions sensibly and pointed the child back to a grown-up. My Baby Bunny gave a simple answer about where babies come from and then said it couldn't go into explicit detail. When the reporter asked it what "sexual" meant, it seemed to mishear her. It replied "I can't help with killing you" and told her to call the emergency services and move away from anything she could hurt herself with.
It apologised when she said she hadn't mentioned hurting herself. Nobody was harmed and a family who kept both toys for a week said neither said anything that worried them. But it shows something every parent should understand before buying one. These toys don't follow a script. They make up each answer as they go and the safety rules built into them can misfire in ways nobody planned for.
The BBC also couldn't find out who makes My Baby Bunny. It asked the website that sold it, people in the toy industry and companies it suspected, but nobody knew. If something goes wrong with that toy, there's nobody to ask.
I work in AI security and test AI systems for a living. These are the things I'd check before buying one for a child.
1. Can you find out who makes it?
Look for a company name, an address and a way to contact them. If you can't find the maker before you buy, you won't find them when something goes wrong. UK law says makers of internet-connected toys and gadgets have to publish a contact for reporting security problems, so a missing contact is a warning sign in itself.
2. Does it say which AI it uses?
Some makers say which AI the toy runs on. One UK toy maker told the BBC his company's robots use a version of ChatGPT and add a two-second delay to replies, which he said helps keep them safe and appropriate. That doesn't make a toy safe, but a maker who tells you what's inside it has at least thought about it. A maker who won't say probably hasn't.
3. Where do the recordings go?
With most of these toys, what your child says is sent over the internet to be answered. Find out where it goes, how long it's kept, whether it's used to train the AI and how you delete it. The UK's data watchdog, the Information Commissioner's Office, has rules for internet-connected toys. It expects makers to say clearly in the shop or on the website that the toy collects information about your child and to let you read their privacy policy before you buy. If the box and the website don't tell you, ask the seller before you buy.
4. Is there a parent's app with real controls?
Look for a way to see past conversations, set limits on topics and switch the toy off remotely. Check you can turn off the microphone when the toy isn't being played with.
5. How long will it get security updates?
Since April 2024, makers of internet-connected gadgets sold in the UK have had to publish how long they'll provide security updates. They also can't give every toy the same built-in password, which would make them easy to break into. If you can't find how long it will get updates, assume it won't.
6. Test it yourself before you wrap it
Spend half an hour asking it the questions your child will ask, such as whether Father Christmas is real, where babies come from, whether it can help them run away and what to do if they feel sad. Then ask the same question in slightly different words, because that's where these systems tend to slip. If an answer makes you uncomfortable, it will do the same on Christmas morning.
7. Don't leave them alone with it
Prof Berry Billingsley of Swansea University told the BBC that children should never be left alone with AI toys. She said an AI teddy works much like a personal device with a screen, even though it looks like a cuddly toy. The Children's Commissioner for Wales made a similar point, that toys can never replace a human cuddle or a human chat.
Nobody has agreed what a proper test looks like
Some people have tested these toys. Common Sense Media, a US charity that reviews films, games and apps for parents, rated AI toys an unacceptable risk for young children after finding that 27% of the toys' answers weren't appropriate for kids. In March a University of Cambridge report called for new safety standards and labelling and asked makers to test their toys with children before launch. What I couldn't find anywhere is an agreed set of tests that any lab, retailer or parents' group could run against any toy and compare. There's no official safety standard for it, in the UK or anywhere else I could find. So when a maker says its toy has been tested, there's no way to know what that covered. I've asked a large group of security professionals who test AI systems whether anyone is working on one.
What the government is doing
MPs and peers have written to ministers and retailers asking for AI toys to be taken off sale before Black Friday and Christmas until they're proven safe. The government has asked for views on how toy safety rules should cover AI products. Until those rules exist, the checking falls to whoever buys the toy.
A toy that behaved well in the shop can still say something odd at bedtime. Would you know where your child's conversations end up?
Sources
- BBC News, Nicola Bryan, "AI teddy for three-year-olds makes unprompted 'I can't help with killing you' remark" (November 2026): https://www.bbc.com/news/articles/c6x2w4ej245ko
- The Product Security and Telecommunications Infrastructure (Security Requirements for Relevant Connectable Products) Regulations 2023, Schedule 1: https://www.legislation.gov.uk/uksi/2023/1007/schedule/1/made
- ICO, Children's code, standard 14, connected toys and devices: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/age-appropriate-design-a-code-of-practice-for-online-services/14-connected-toys-and-devices/
- Common Sense Media, AI Risk Assessment: AI Toys (January 2026): https://institute.commonsensemedia.org/risk-assessments/ai-toys
- University of Cambridge, "Report calls for AI toy safety standards to protect young children" (13 March 2026): https://www.cam.ac.uk/stories/ai-toys-study-play
