AI toys for children: what to check before one comes into the house

Name: AI toys for children: what to check before one comes into the house
Item: AI companion toys (category assessment)
Author: DailyTech Staff

Talking companions with a language model inside are now a shelf category. The differences that matter are not the ones on the box.

DailyTech Staff · 1 September 2026 · 7 min read

The verdict

AI toys differ from ordinary toys in three ways that matter: they listen continuously, they send what they hear somewhere, and they stop working when the company behind them stops paying for servers. The right questions before buying are where the audio goes, what happens if the service shuts down, and whether the conversation is bounded. A toy that answers all three clearly is a reasonable purchase; most of the failures in this category so far have been companies unable to answer any of them.

Buy it if

Households happy to treat the toy as an internet-connected device with a microphone, and to supervise it the way they would a tablet.

Skip it if

You want an offline toy, you are not prepared to manage an account and a subscription, or the child is young enough that an open-ended conversational partner is not appropriate.

Strengths

Conversational play holds attention far longer than fixed-script talking toys
Language practice and question-answering can genuinely be valuable for older children
The better products publish their content boundaries and parental controls plainly

Limitations

Always-listening microphones in a child's room are a category of risk, not a feature detail
Several high-profile AI companion toys have shut down and bricked, taking the device with them
Safety filtering on open-ended models is imperfect; reported failures in this category are real
Subscription dependence means the toy's lifespan is a business decision, not a durability one

Key facts

Price: Roughly $50 to $400 depending on the device, with several requiring subscriptions
Typical connectivity: Wi-Fi, with speech processed in the cloud rather than on the toy
Typical dependency: Vendor account, companion app, and often a monthly or annual plan
What to check: Audio retention policy, deletion controls, and whether recordings train models
Failure mode: Service shutdown renders most of these devices non-functional

How we assessed this: Category assessment built from manufacturers' published documentation, consumer-safety reporting and the public record of products in this category, including ones that have been withdrawn. Individual toys were not tested by us.

A talking toy used to contain a chip with a fixed set of phrases. A modern AI toy contains a microphone, a network connection and a link to a language model somewhere else. The play value is real — children will talk to a responsive companion for far longer than a scripted one — but the object on the shelf is closer to a smart speaker than to a teddy bear, and it should be evaluated that way.

Question one: where does the audio go

Almost all of these devices process speech in the cloud. The questions worth asking are how long the recordings are kept, whether a parent can delete them, whether they are used to train models, and who else can access them. Manufacturers that answer these clearly in plain language on their website tend to be the ones who have thought about it. Manufacturers who bury it are telling you something.

A related practical point: the toy hears everything in the room, not only the child. Family conversation, other people's children, whatever is on television.

Question two: what happens when the company stops

This category has already produced shutdowns in which a well-reviewed, well-funded companion robot stopped working because the company behind it ran out of money. Families were left with an inert plastic object and a child who had formed an attachment to it. That is a foreseeable outcome for any device whose intelligence lives on someone else's server.

Before buying, look for whether the maker commits to any offline mode or refund policy on shutdown. Very few do. Price the toy as a subscription with a physical shell rather than as a durable good, and set expectations with the child accordingly.

Question three: is the conversation bounded

A general-purpose language model in a toy will, given enough conversation, be steered somewhere it should not go. Products that restrict the model to a defined persona and subject range, publish their content policy and provide parental review of transcripts are meaningfully safer than products that put a general assistant into a plush body. Consumer-safety groups testing this category have repeatedly found inappropriate responses in the second kind.

Parental controls that exist are not the same as parental controls that are on by default. Check both.

The practical recommendation

Treat an AI toy exactly as you would a tablet: it lives in shared space rather than the bedroom, the account is a parent's, the transcript gets looked at occasionally, and the child knows it sends what it hears to a company. Under those conditions the better products in the category are fine and some are genuinely good.

The one purchase to avoid is the cheap unbranded device with no published privacy policy and no named company behind it. There are a lot of them, they are the ones most likely to be listening carelessly, and they are the ones most likely to be gone next year.