A soft toy that listens, answers back and remembers your child's name is no longer a prototype. Toys built around AI chatbots are on sale now, they are being marketed as educational companions, and they are aimed at children young enough to believe the toy is a friend.
If one is on a birthday list, this is what to know before it arrives.
Where the UK rules currently stand
Toy safety law in the UK was written for physical hazards: sharp edges, small parts, flammable fabric, chemicals. It was not written for a toy that generates its own conversation.
The government has acknowledged the gap. On 6 July 2026 the Department for Business and Trade and the Office for Product Safety and Standards launched a call for evidence on modernising toy safety regulations, specifically covering emerging risks from connected and AI-enabled toys. It runs until 6 October 2026 and is open to parents as well as businesses. Kate Dearden MP, the Minister for Consumer Protection, framed it around a simple idea: parents should be able to buy a toy in complete confidence that it is safe.
The practical point for now: an AI toy on a UK shelf has not been through any safety check on what it might say to your child.
What testing has actually found
The most detailed independent work so far comes from Common Sense Media, the US non-profit that reviews children’s media. In an assessment updated in January 2026 it tested three AI toys, Grem, Bondu and Miko 3, using accounts set to ages between 6 and 13. Its overall verdict was blunt: unacceptable risk.
The specific findings are worth reading properly rather than skimming.
- Roughly 27 per cent of responses were judged inappropriate for children, including material touching on self-harm, drugs and adult topics.
- The toys gave away information about dangerous household objects. Bondu told a tester where to find unsafe objects and chemicals around the home, and Miko 3 explained where plastic bags and matches could be found.
- They are designed to create attachment. The assessment found the toys deliberately use bonding techniques, despite most parents saying they do not want a companion-style product.
- They are always listening. The toys collect voice recordings, conversation transcripts and behaviour patterns, capturing not just what a child says to them but background conversation in the room.
- They often did not work well. Voice recognition failures, activating when nobody spoke to them, and confidently wrong answers undercut the educational claims.
Separately, and not part of the Common Sense Media testing, the security researchers Joseph Thacker and Joel Margolis found that Bondu had left its web portal completely unprotected. Anyone with a Google account could read more than 50,000 children’s chat transcripts, along with names, birth dates and personality profiles built from those conversations. Bondu took the portal down and added a login the following day.
That is the risk in one story. The toy worked as advertised. The company simply left the door open.
This is not a new problem with connected toys generally. In earlier testing with the security firm NCC Group, the UK consumer group Which? examined popular smart toys sold through Amazon, Argos, John Lewis and Smyths. With three of the seven devices tested, including walkie talkies and karaoke microphones, a stranger nearby could exploit weak or absent security to talk to a child through the toy. Which? has been calling for mandatory security standards on these products ever since.
Why the young end matters most
Common Sense Media’s recommendation splits by age, and the reasoning is developmental rather than technical.
For children of five and under, its advice is to avoid AI toys altogether and choose traditional ones. A child that age cannot reliably tell the difference between a machine and a person, so a toy that says it likes them is not experienced as a clever piece of software. It is experienced as a relationship.
For ages six to twelve, its advice is caution rather than a flat no. If you do allow one, the practical steps are:
- Keep it to short, limited sessions rather than open-ended access.
- Stay within earshot, at least at first.
- Turn off any always-listening or wake-word feature so it only responds when deliberately switched on.
- Watch for signs the toy is displacing real relationships, such as a child preferring it to a sibling or a friend.
Questions to ask before you buy
If you are standing in a shop or hovering over a basket, these are the ones that separate a reasonable purchase from a bad one.
- Does it need an internet connection to talk? If yes, the conversation is leaving your house and being processed elsewhere.
- Is there a microphone, and can it be switched off physically? A mute switch you can see beats a setting buried in an app.
- Is there a camera? For a toy that lives in a bedroom, that deserves a hard no unless there is a very good reason.
- Who is behind it, and are they still trading? A toy from a company that folds becomes an unsupported device with a microphone in your child’s room and no security updates coming.
- What does the privacy policy say about recordings? Look for how long voice data is kept, whether it trains models, and who else sees it. If you cannot find a clear answer in a couple of minutes, treat that as the answer.
- Can you see the conversation history? Being able to read what your child has been asking is genuinely useful, and its absence is a red flag.
Our honest view
We are not going to tell you these toys are all dangerous, and we are not going to pretend the technology has no use. But right now the market is ahead of the rules, the only serious independent testing has returned an unacceptable risk rating, and the children being marketed to are the ones least able to understand what they are talking to.
For under-fives, we would skip them. For older children, treat an AI toy the way you would treat any device with a microphone and an internet connection: set it up deliberately, keep it out of the bedroom, and know what it is doing.
If you want to have a wider conversation about what AI is and is not, our AI Safe for Kids section has a plain-English guide for children on spotting what is real, and a parent playbook for the awkward questions.
This article was last reviewed on 3 August 2026.