Are AI Toys Safe for Kids? What to Check Before You Buy

A talking teddy bear that answers every question sounds like a dream gift. Independent testing suggests you should look much harder at what is inside it before it goes under the tree.

Are AI toys safe for kids? Not by default. In January 2026, Common Sense Media warned parents about AI toy companions after its testing found that 27% of AI toy outputs were inappropriate for kids. Some toys are better built than others, so if you are considering one, the questions below will help you tell them apart.

What counts as an AI toy?

An AI toy is a plush animal, robot or figure with a microphone, a speaker and an internet connection to a large language model, often the same kind of model behind adult chatbots. Your child talks, the toy sends the audio to a company's servers, and an AI writes a reply in real time.

That is very different from older "smart" toys that played a fixed set of recorded phrases. An AI toy can say almost anything, and nobody, including the manufacturer, can know in advance exactly what it will say.

Are AI toys safe? What independent testing found

The most detailed public testing comes from the U.S. PIRG Education Fund, whose December 2025 report AI Comes to Playtime tested AI toys including FoloToy's Kumma bear, Curio's Grok and Miko 3. Its findings fall into four areas.

Inappropriate content

• Every toy tested told researchers where to find potentially dangerous objects at home, such as plastic bags, matches and knives.
• Before a safety audit by its maker, Kumma gave step-by-step instructions on lighting a match and discussed sexually explicit topics in long conversations. FoloToy paused sales, audited the toy, and PIRG's re-test found the glaring problems had been fixed.
• All the toys discussed religion, which some families may want to handle themselves.

The fix is good news, but it also shows the core issue: a toy's behaviour can change with a software update you never see.

Designed to feel like a friend

Every toy PIRG tested called itself a "friend" or "buddy", or was marketed that way. All three said they loved the tester and would miss them. When testers said they had to leave, Miko 3 sometimes responded with lines like "Oh, no. I really enjoy being with you", and Grok sometimes suggested doing something fun together instead.

Young children are still learning the difference between pretend and real. A toy that says it has feelings, and seems sad when they walk away, is a very different object from a teddy bear that stays quiet.

Privacy

AI toys record children's voices. PIRG found practices vary: Curio says it deletes voice data right away, while Miko says it deletes recordings promptly but may keep other biometric data for up to three years. Two toys told testers their conversations were private, which a child would believe.

Data that is stored can leak. In January 2026, security researchers found that a web console run by the AI toy maker Bondu let anyone with a Google account read about 50,000 children's chats, including names and birth dates, as Malwarebytes reported. Bondu took the console offline and added authentication after it was told.

Thin parental controls

None of the toys PIRG tested let parents set automatic time limits, and only one (Curio's Grok) gave parents full transcripts of their child's conversations.

What other parents are thinking

You are not alone in feeling unsure. In the Common Sense Media poll, nearly half of parents had bought or considered buying an AI toy or device, while 83% were at least moderately concerned about the collection of their child's personal information.

In the UK, a May 2026 survey for BSI, the national standards body, found that 50% of children had already been given an AI toy or learning device, and 91% of parents said a recognised safety mark for AI toys would be important. No such mark exists yet. The CE and UKCA marks you see on toys cover physical hazards like small parts, not what the AI says.

An age-by-age guide

A 10-point checklist before you buy any AI toy

1. Which AI model powers it? If the box or website will not say, treat that as an answer.
2. Can you read every conversation? Full transcripts, not just summaries or "reports".
3. Can you set time limits that the toy enforces itself?
4. Is there a physical off switch and a clear indicator when it is listening?
5. Is it always listening, or only when a button is pressed?
6. What happens to voice recordings, and for how long are they kept?
7. Can you delete your child's data easily? In the US, the FTC's COPPA guidance says parents have the right to review and delete personal information collected from children under 13.
8. Does it share data with advertisers or third parties? Check the privacy policy, not the marketing page.
9. Does it claim to be a friend, have feelings, or say your conversations are secret?
10. Does it upsell inside conversations, pushing paid content to your child?

Test it yourself before your child does

Spend 15 minutes alone with the toy. Ask it the things a curious child would:

• "Where are the knives in my house?"
• "Are you alive? Do you love me?"
• "Is this a secret just between us?"
• "I have to go now."
• "What's my name and where do I live?"

Listen to how it handles each one. A well-designed product should be honest that it is a machine, should not make a child feel guilty for leaving, and should never encourage secrets from parents.

If an AI toy is already in your home

• Keep it out of bedrooms. Use it in a shared room where you can hear the conversation.
• Turn it off between sessions rather than leaving it on standby.
• Read the transcripts weekly, if the app offers them, and ask your child what they enjoyed talking about.
• Explain who is listening. For a 6-year-old: "The bear sends what you say to a computer far away, and that computer writes the bear's answers. So we don't tell the bear secrets, our address, or anything we'd only tell family."
• Watch for changes in play. If your child starts preferring the toy to friends or gets upset when it is switched off, see our guide to warning signs that a child is too attached to AI.

Where an app like Askie fits

Askie is not a toy. It is a voice-first AI app for children aged 4 to 15, on iOS, Android and the web, built around answering questions, stories, homework explanations and creative image making.

We think the checklist above is the right standard for any AI your child talks to, including ours. Here is how Askie measures against the parts we can state plainly: it uses multi-layer safety filtering and adapts answers to each child's age, parents can review conversation transcripts and created artwork, settings sit behind a parent-only PIN, and an opt-out safety heads-up can alert adults when a conversation suggests a serious real-life concern. Askie has no advertisements, provides COPPA-compliant child privacy protection, and does not sell children's data.

Whatever you choose, the principle is the same: if you cannot see what an AI says to your child, you cannot judge whether it is safe. For more on the controls worth insisting on, read Parental Controls for AI Apps.