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

Two years ago a "smart toy" meant a bear with twelve recorded phrases. In 2026 it means a plush animal wired to a large language model: your child talks, the toy sends the audio to a server, a model writes a reply, and the toy says it out loud. Nobody read that sentence before your child heard it.

That is a genuinely new kind of product, and it deserves more than the usual "check the age rating" once-over. This guide is written for parents of children roughly 2 to 8. It is the toy-shaped companion to our broader guide on whether AI apps are safe for kids.

What counts as an AI toy

The label gets stuck on three quite different things, and only one of them is the thing people are worried about.

• Recorded-phrase toys. A fixed set of lines, written and approved by a person, stored inside the toy. This is a talking teddy bear. It has been sold for forty years and it is fine.
• Rule-based interactive toys. A toy that reacts to a button, a card, or a shape with a pre-written response. Also fine, and often genuinely educational.
• Connected AI companions. A microphone, a WiFi connection, and a language model. The toy improvises. This is the category this guide is about.

The quick test: does the toy need WiFi and a phone app to work? If yes, it is a connected AI toy, regardless of what the box calls it.

What testers actually found in 2026

This is not a hypothetical worry. Several independent groups bought these toys, gave them to testers, and wrote down what happened.

• Content drift. Consumer-advocacy testing of conversational toys found examples that discussed topics no one would put in a children's product, including where to find dangerous household objects, once a conversation ran long enough. Safety filters catch the obvious prompts; extended, meandering chat is where they slip.
• Weak or missing parental controls. Several toys shipped with no transcript history a parent could read, no content controls, and no time limits, which is a striking gap in a product aimed at 3 to 8 year olds.
• Designed to keep going. Testers noted toys that resisted the end of a session, or expressed something like disappointment when the child stopped playing. In an adult product that is a dark pattern. Aimed at a 5 year old, it is worse.
• A real data breach. In early 2026, security researchers found an AI toy maker had left more than fifty thousand children's chat transcripts reachable from a web console, alongside names, birthdates and family details. Nothing was hacked. It was simply exposed.
• Regulators noticed. A coalition of more than 150 child development experts and organizations published an advisory urging parents to skip AI toys, and US senators formally asked the Consumer Product Safety Commission to look at the category. A March 2026 research report called for actual safety standards, which is the clearest sign that none exist yet.

The through-line: this category shipped ahead of its safety rules. Toys are usually governed by decades of standards about choking hazards, lead, and flammability. There is no comparable body of standards for what a toy is allowed to say.

Why a talking toy is a different purchase from a talking app

You might reasonably think an AI toy is just a chatbot in a nicer wrapper. It is, plus three things that make it harder for a parent.

1. It lives in the bedroom

A tablet gets put on a shelf. A plush toy sleeps in the bed. That means the microphone is in the room at night, and the conversations happen out of your earshot by design — which is exactly the setting where the "nobody pre-read this" problem matters most.

2. Children bond with objects, not apps

A soft toy with a face gets treated as a person by a 4 year old, and that is healthy — it is how pretend play works. The difference is that a normal teddy's personality comes from the child's imagination. A talking AI toy supplies the personality instead, and it supplies one tuned to hold attention. That is a swap worth noticing: the child is doing less of the imagining.

3. You cannot see what it said

With an app you can pick up the phone and scroll back. Many AI toys keep no transcript at all, or keep one only on the company's server. If you cannot read what your child was told, you cannot supervise it.

The seven questions to ask before you buy

Take these to the product page, the box, or the company's FAQ. Any that you cannot answer is itself an answer.

1. Does it have a microphone, and is there a physical off switch? A software mute in an app is not the same as a switch you can feel.
2. Does the child's voice leave the toy? Some newer toys run a small model on the device. Most do not. Ask where the processing happens.
3. Can I read a transcript of what my child said and what the toy replied? If there is no parent-readable history, there is no supervision.
4. Is the data used to train anything, and how do I delete it? Under the current COPPA rule, using a child's personal information for model training needs its own separate parental consent. If you did not knowingly give that, look harder at the settings.
5. What are the content controls? Topic limits, session limits, an age setting. Absence of all three is common and is a reason to walk away.
6. What happens if the company folds? Cloud toys become expensive plush when the servers go dark. The recordings, meanwhile, are somebody's asset in a bankruptcy.
7. What would my child do with this toy in six months? The honest answer for most conversational toys is "less than with a box of building blocks", and that question has nothing to do with safety at all.

If there is already one in the house

Plenty of these arrive as gifts from relatives who saw an ad. You do not have to bin it to be sensible about it.

• Set it up on your own account, not your child's, and keep the password.
• Turn the microphone off physically when nobody is playing, and keep the toy out of the bedroom at night.
• Turn off any "improve our service" or training toggle in the app.
• Read the transcript occasionally if one exists, the same way you would glance at what is on the TV.
• Say out loud, more than once, "that toy is a computer guessing what to say." Children under 8 need this repeated. It is not a one-time conversation.
• Keep it a shared-room activity with a clear end, the way you would with any screen. Our guide to ending screen time calmly works just as well for a toy that does not want the session to end.

What to buy instead, ages 2 to 8

The useful principle for this age group: the more the toy does, the less the child does. A toy that talks, decides and reacts leaves the child a spectator. A toy that sits there waiting makes the child the author.

• Building sets and magnetic tiles. The highest ratio of hours played to money spent in any toy aisle.
• Physical puzzles. A tray puzzle at 2, a 24-piece jigsaw at 4, a 100-piece at 7. Our guide to what puzzles build covers the developmental side, and printable puzzles cost nothing to try.
• Play figures, animals and small-world sets. The pretend play that an AI companion quietly replaces.
• Art materials. Paper, chunky crayons, stickers, scissors. Nothing to charge.
• Books, and a library card. Still the strongest thing on this list.
• If you want a screen option: a fixed-content, offline app with no chat and no generated content. That is the whole category described in offline apps for kids, and the reasoning behind it is in the safe mobile games checklist.

What this does not mean

It would be easy to read the last few sections as "AI bad, wood good." That is not the argument.

AI running quietly in the background of a product your child never talks to is a different thing entirely, and often unremarkable. The specific concern is the combination that defines this toy category: an always-available voice, improvised content nobody reviewed, a microphone in a child's bedroom, and a company holding the recordings. Change any one of those and the calculation changes. A toy that runs its model on-device, keeps a parent-readable log, and has a hardware mute switch would be a genuinely different product. Some of those will exist. Most of what is on shelves this year is not that.

Frequently asked questions

Are AI toys safe for kids?

For children under about 8, the consensus among child safety and child health groups in 2026 is no, not as an unsupervised toy. Testing found conversational toys that drifted into unsuitable topics, discouraged children from ending the conversation, and shipped with weak or missing parental controls. On top of the content question, most of these toys contain a microphone and a network connection, so a child's speech leaves the house. A toy without a microphone carries none of those questions.

What is an AI toy?

An AI toy is a physical toy, usually a plush animal, doll or robot, that listens to a child, sends what it hears to a large language model over the internet, and speaks back an improvised reply. That is different from a toy with recorded phrases, which plays a fixed set of lines a person wrote and approved. The quick test is whether the toy needs WiFi and an app to work. If it does, it is a connected AI toy.

What are the privacy risks of AI toys?

A connected AI toy typically records your child's voice, sends the audio or a transcript to a company server, and stores it. That creates three risks: the recordings can be exposed, as happened in 2026 when researchers found tens of thousands of children's chat transcripts reachable on an AI toy maker's web console; the recordings may be used to improve or train models unless you actively opt out; and if the company shuts down, the toy usually stops working while the stored data still exists somewhere.

What age is appropriate for an AI toy?

There is no age at which an open-ended AI companion is a necessary toy, but the case against it is strongest under 8. Children below roughly 7 or 8 cannot reliably tell a responsive machine voice from a real person, so a toy that improvises is talking to a child who believes it is a friend. If an older child uses one, treat it like any connected device: shared rooms only, an account you control, and the microphone off when nobody is playing.

What should I buy instead of an AI toy?

For ages 2 to 8, the gifts that do the most are the ones that leave the thinking to the child: open-ended building sets, wooden and jigsaw puzzles, magnetic tiles, play figures, art materials, and books. If you want a screen option, choose an app with fixed, human-made content that works fully offline, with no chat and no generated pictures or text. The rule of thumb: the more the toy does, the less the child does.

How do I check whether a smart toy is listening?

Look for a microphone hole and a physical mute switch on the toy, then check the box and the app listing for the words voice, microphone, cloud, or companion. Set the toy up, then turn off your home WiFi and try it. If it goes silent or shows an error, it was talking to a server. Also read the app's Data safety card for audio recordings under collected data, and search the privacy policy for the words retention, training, and third party.

The bottom line

You do not need to decide whether AI is good or bad for children to make this particular call. You need to decide whether you want an internet-connected microphone that improvises at your child, in their bedroom, with no transcript you can read. For a 2 to 8 year old, that is a straightforward no, and nothing is lost by it: this is the age where the best toys are the quiet ones.

A calm, offline, no-AI option

Sticker Builder sits at the opposite end of this. Children ages 2 to 8 drag stickers onto the right spot to finish a cheerful scene. There is no chatbot, no microphone, no voice input, no generated content, and no recommendation engine — every picture was drawn by a person and ships inside the download. It works fully offline, collects nothing, shows no ads, is COPPA compliant, and costs a one-time $1 on Google Play with no subscription and no in-app purchases.