Safety checklist

Are AI toys safe for kids?

By Frederik Feldt, founder of SkyBuds. Updated October 29, 2026.

The short answer: an AI toy can be safe for kids if you can check five things: when it listens, what it keeps and for how long, whose AI answers your child, whether you can read the conversations, and whether it passed toy safety tests. No AI filters perfectly, so a parent app you actually read is the best safety net.

Below is the checklist we would use on any AI toy, ours included, with SkyBuds’ answer to each point. We make SkyBuds, so read our answers with that in mind, and ask every other maker the same questions. If a company cannot answer them clearly on its own site, that is an answer too.

A girl at a table holding Sprout, the green SkyBuds plush bear, while her dad shows her an autumn leaf
Sprout at home. A SkyBud has no camera and no screen.

Why parents are asking

AI toys now get real scrutiny. In December 2025, Senators Marsha Blackburn and Richard Blumenthal demanded answers from six AI toy makers, Miko, Curio and KEYi Robot (which makes Loona) among them (NBC News). In February 2026 the senators said Miko had left what appeared to be the toy’s audio responses to children in a publicly accessible database; it was taken offline, and Miko’s CEO told NBC News there had been no breach or leak of user data (NBC News).

And Moxie, a robot that needed its maker’s servers, is no longer for sale; its makers say it is being sunset (moxierobots.com; what happened to Moxie).

The AI toy safety checklist

AskWhy it mattersSkyBuds’ answer
When does it listen?A toy can listen on a button press, on a wake word, or all the time.SkyBuds is not always listening. A small chip on the toy listens only for the wake word “Hey SkyBud”, and only while the toy is in standby. When the toy is powered off, nothing listens.
Is there a camera or a screen?Each one is more data and more to supervise.No camera. No screen. No ads.
Whose AI answers your child?Your child’s words go to whichever company runs the AI.The AI your child talks to is SkyBuds’ own model, running on our own servers. It is not ChatGPT, Claude, Gemini or any other AI lab’s model. Speech recognition, the memory model that works out what to remember, and the safety checks run on those same servers, so when your child talks with their SkyBud, their voice is not sent to any outside AI company’s model. Only the toy’s own reply text goes to our voice provider, under a zero-data-retention agreement.
What is kept, and for how long?Conversations and recordings are the most personal data a toy holds.We keep conversation transcripts for 12 months so the SkyBud remembers and you can read along in the app; then they are deleted. Keeping some voice recordings to train and test SkyBuds’ own speech models is part of the permission you give at setup; you can turn it off at any time in the app, and the toy works exactly the same either way. Every kept recording is deleted within 24 months. What SkyBuds records and for how long.
Can you read the conversations?This is how you catch what a filter misses.Yes. The free parent app shows a short summary and the full transcript of every chat, and worrying moments get flagged so you know when a caring check-in may help.
Who set it up?A child should not be able to switch an AI toy on for themselves the first time.A parent or legal guardian sets up and approves every SkyBud before a child uses it.
Is children’s data sold?Ask it plainly.No. We don’t sell children’s data. You can delete your account and all of your child’s data, anytime.
Is it tested as a toy?An AI toy still has a battery, seams and small parts.Tested by an independent, CPSC-accepted lab to CPSIA and ASTM F963-23, and to the EU toy standards EN 71 and EN IEC 62115. Complies with Part 15 of the FCC Rules. The battery is sealed in and cannot be removed.
Does an independent program vouch for it?A seal you can check on the program’s own site is worth more than a badge.SkyBuds is kidSAFE Listed, a seal from the independent kidSAFE Seal Program. COPPA compliance work is in progress and not yet a completed certification.
What if the company stops?A toy whose AI runs on its maker’s servers depends on that maker.Our privacy policy promises that if Firefly Jar ever winds down the SkyBuds service, we will give you advance notice by email, delete children’s personal information, and publish a plan for the toys themselves (section 11).

Which AI toys have parental controls?

The controls live in a parent app, and apps do different things, so ask what you can actually see and change. With SkyBuds, the free parent app lets you:

  • Read every chat: a short summary and the full transcript.
  • Get worrying moments flagged in the app, so you know when a caring check-in may help.
  • See what your child’s SkyBud remembers.
  • See your week: talk time, streaks and top topics.
  • Set quiet hours: the SkyBud sleeps at the times you pick.
  • Delete your account and all of your child’s data, anytime.

What other makers say on their own pages, checked October 29, 2026:

  • Ted: you choose push-to-talk or auto-listen in the parent app (Ted&Co’s FAQ), and its site lists quiet hours and parent alerts (tedco.ai).
  • bondu: parents are notified in its Parent App if a concerning topic comes up (bondu’s FAQ).
  • TalkiPal: parents stay connected and in control through its app (talkipal.com).
  • Miko: parents can disable Miko 3’s camera or mic with one tap (miko.ai).

Those are the makers’ own descriptions; none of them is a test we ran. Before you buy, ask whether you can read the full conversations, not only alerts or summaries.

Is there a safe AI toy for a 5-year-old?

The checklist is the same at 5 as at any age; what changes is which toys are made for your child. Age ranges differ a lot. SkyBuds is made for ages 3 to 10. On their own pages, checked October 29, 2026: Curio says primarily 3 to 12 (Curio’s FAQ), Ted from 3 (Ted&Co’s FAQ), TalkiPal above 3 (talkipal.com), bondu 4 to 8 (bondu’s FAQ), Poe the AI Story Bear 4 and up (TTPM), and Miko says its AI conversations are designed for kids 5 to 10 (miko.ai).

A 5-year-old is inside every one of those ranges, so the age on the box will not decide it for you. Pick a toy whose maker answers every question in the checklist on its own site, check whether it has a camera or a screen you would rather avoid, and read the chats for the first few weeks. A SkyBud adapts its language, topics and complexity to your child’s age, so it changes with them as they grow.

For a younger child, see the best AI toy for a 3-year-old; for every toy side by side, the 2027 AI toy guide.

Where to read more

More guides for parents

Questions parents ask

In short.

Are AI toys safe for kids?

They can be, if you can check when the toy listens, what it keeps, whose AI answers your child, whether you can read the chats, and whether it passed toy safety tests. No AI filters perfectly, so a parent app you read is the best safety net.

Is SkyBuds always listening?

No. A small chip on the toy listens only for the wake word “Hey SkyBud”, and only while the toy is in standby. When the toy is powered off, nothing listens.

Does SkyBuds use ChatGPT?

No. The AI your child talks to is SkyBuds’ own model, running on our own servers. It is not ChatGPT, Claude, Gemini or any other AI lab’s model.

What parental controls does SkyBuds have?

In the free parent app you can read every chat, get worrying moments flagged, see what the SkyBud remembers, see your week, set quiet hours, and delete your account and all of your child’s data, anytime.

Is SkyBuds safe for a 5-year-old?

SkyBuds is made for ages 3 to 10, so a 5-year-old is in its range. It is kidSAFE Listed, has no camera, no screen and no ads, and was tested by an independent, CPSC-accepted lab to CPSIA and ASTM F963-23.

Meet the SkyBuds.

Find your SkyBud
  • 30-day money-back guarantee
  • Talking and stories free for life