Table of Contents
AI Toy Data Leak: The Plush That Exposed 50,000 Kids' Chats
An AI toy data leak let anyone with a Gmail account read about 50,000 kids' chats. How cloud toys store chats, how the console failed, and what to ask.
No hacking was needed. That is the line in the report that should stop every parent who owns a talking toy.
In early 2026, security researchers found that Bondu, a company selling an AI plush toy, had a web console where anyone could log in with any Gmail account and read children’s conversations with the toy. About 50,000 chats were exposed, including kids’ names, birth dates, family details and intimate conversations, according to Malwarebytes’ February 3, 2026 report. Bondu closed it within minutes of being told. But this AI toy data leak is worth understanding in detail, because the design that made it possible is common across cloud-connected toys.
Key Takeaways
- Bondu’s console let anyone log in with an arbitrary Gmail account and read stored chat transcripts. About 50,000 conversations, covering virtually every child who used the toy, were visible.
- The flaw was an authorization failure: the system checked identity (a valid Google login) but not permission (is this person a parent of this child, or staff?).
- A week later, two U.S. senators said Miko had left what appeared to be all of its toy’s audio responses in an unsecured database, some naming children. Miko’s CEO said there was “no breach or leak of user data.”
- Common Sense Media’s 2026 testing found AI toys collect voice recordings, transcripts and behavioral data, and that even maximum privacy settings still require significant collection.
- Before buying, ask four questions: what is stored, where, for how long, and who can see it.
The Bondu AI Toy Data Leak: What Happened
Bondu markets its toy as “a soft, cuddly toy powered by AI that can chat, teach, and play with your child.” Like other AI toys, it records what a child says, sends it to the company’s servers, generates a reply with a chatbot and plays it back. Conversations are stored so the toy can remember the child and so parents can review them.
According to Malwarebytes’ report by Pieter Arntz, two researchers (whose account was first reported by Wired) found that Bondu’s public-facing web console accepted a login from any Gmail account. Once in, they could read chat transcripts. The only conversations not available were those parents or staff had manually deleted.
Bondu took the console offline within minutes of disclosure and later relaunched it with proper authentication. Its CEO said fixes were completed within hours, that the company found “no evidence” of other access, and that it had brought in a security firm and added monitoring. Those are the company’s statements. “No evidence” of access is not the same as proof that no one looked, which is often impossible to establish after the fact.
How a Cloud-Connected Toy Stores Conversations, and Where It Broke
An AI toy’s conversations pass through several places, and each is a place data can sit.
- The toy. A microphone captures audio when the toy is listening. Some toys use a push-to-talk button, some a wake word, and some listen continuously, according to U.S. PIRG’s 2025 testing.
- The company’s cloud. Audio is sent over Wi-Fi to servers, where speech recognition turns it into text. The text goes to a large language model, which writes a reply. Text-to-speech turns the reply into audio.
- Storage. Transcripts, and sometimes audio, are saved in a database. This is what lets the toy “remember” your child’s favorite color, and what lets the company improve the product.
- The console. A website or app lets parents (and company staff) view those stored conversations.
The Bondu failure happened at step 4. In engineering terms, there are two separate questions any secure system must ask:
- Authentication: Are you who you say you are? A Google login answers this.
- Authorization: Given who you are, are you allowed to see this record? Only the toy’s records, and only your child’s.
Bondu’s console, as described in the reports, did the first and skipped the second. Imagine a building whose guard checks that you have a valid driver’s license and then hands you keys to every apartment. The ID is real. The access is wrong.
The Miko case: a different hole, the same lesson
On February 12, 2026, NBC News reported that Sens. Marsha Blackburn and Richard Blumenthal had written to Miko, saying the company left “what appears to be all of the audio responses of the toy” in an unsecured, publicly accessible database. Senate staff found it using free tools to examine the toy’s Wi-Fi traffic. NBC found files dating back to December 2025, organized by date, some linkable to a named child over several minutes, but no recordings of children’s own voices. The index contained folders labeled “GOOGLE” and “AZURE,” the names of two major cloud providers.
Miko’s founder and CEO, Sneh Vaswani, said, “There has been no breach or leak of user data,” and that “Miko does not store children’s voice recordings.” The database was taken private after the senators contacted the company.
Notice what leaked here: only the toy’s side of the conversation. It still revealed names, timing and what children had talked about. A toy’s replies are a mirror of your child’s words.
This isn’t only a small-company problem
In May 2023, the Federal Trade Commission and Justice Department charged Amazon with violating children’s privacy law by keeping kids’ Alexa voice recordings indefinitely unless parents asked for deletion, and with failing to fully delete transcripts when they did. Amazon agreed to a $25 million civil penalty, according to the FTC. Retention is risk, whatever the size of the company.
How to Teach Your Kid About AI Toy Data
Ages 5–8: The shoebox server
Take a shoebox and label it “Toy Company.” Each time your child tells a stuffed animal something, write it on a slip and drop it in the box. After five slips, ask: “Who can open this box?” Then add a lock (a rubber band) and a key (a paper clip) and decide together who gets a key. The lesson: what you tell a talking toy goes into a box somewhere else, and the important question is who has the key.
Ages 9–12: Bouncer vs. guest list
Set up a pretend party at the kitchen door. One person is the bouncer who checks ID (“Show me your library card”). The other holds the guest list. Run it twice: once with only the bouncer, where anyone with a card gets in, and once with both, where you need a card and your name on the list. Then explain that the Bondu console had a bouncer but no guest list. Ask your child to say, in their own words, the difference between proving who you are and being allowed in.
Ages 13+: Map one device’s data flow
Pick any connected device in your home: a smart speaker, a game console, a toy. Have your teen draw the path of one sentence spoken to it, from microphone to cloud and back, labeling every place it might be stored. Then find the device’s privacy policy and answer four questions: What is collected? How long is it kept? Who else gets it? How do you delete it? Most teens will find at least one question the policy doesn’t clearly answer. That gap is the finding.
The question to ask: “If a toy company checks that you have a real email account, why isn’t that enough to keep strangers out of your conversations?”
Where a Talking Toy’s Data Goes, and What Can Go Wrong
| Stage | What’s there | Who can usually see it | Real failure |
|---|---|---|---|
| Toy microphone | Your child’s voice, sometimes background talk | No one, until sent | Always-listening toys capture family conversations (Common Sense, PIRG) |
| Network traffic | Audio and data in transit | Anyone examining the toy’s traffic, if not protected | Senate staff found Miko’s exposed files by examining Wi-Fi traffic |
| Cloud processing | Speech-to-text, chatbot, text-to-speech | Company and its cloud vendors | Miko files sat in folders named for Google and Azure |
| Database | Transcripts, audio, profile (name, birthday) | Company staff | Amazon kept kids’ Alexa recordings indefinitely (FTC, 2023) |
| Parent/staff console | Readable conversation history | Should be: parent and authorized staff | Bondu: anyone with any Gmail account |
What to Do Before (and After) You Buy a Talking Toy
Ask the four storage questions
What does it store (audio, transcripts, or both)? Where (which company, which country)? For how long? Who can see it? If the box, website and privacy policy can’t answer all four plainly, treat that as the answer.
Prefer toys that keep less
Push-to-talk beats wake word, and wake word beats always-listening. A toy that doesn’t keep transcripts can’t leak them. Common Sense found that even with maximum privacy settings, AI toys still require significant data collection to work.
Delete on a schedule
If the toy has a parent app with conversation history, delete it regularly. Bondu’s exposed set excluded only conversations that had been manually deleted. That is the one control that worked.
Know your rights
In the U.S., the Children’s Online Privacy Protection Act gives parents rights over data collected from children under 13. Our guide to what companies owe kids under COPPA covers how to request deletion. For broader privacy checks, see AI smart toys and kids’ privacy.
What not to do
Don’t try to “test” a toy’s console yourself by poking at other accounts. Accessing data you aren’t authorized to see can be illegal, even if a system lets you. If you find a hole, report it to the company and, if it’s serious, to the FTC.
What to Watch For Over the Next 3 Months
- Week 4: Open the toy’s app and look at what it has stored about your child. Is there a transcript history? A profile with a birthday? Delete what you don’t want kept.
- Month 2 red flags: The company changes its privacy policy, adds third-party sharing, or you receive a breach notice. Any of these is a reason to stop using the toy’s cloud features.
- Month 3 self-check: Could you answer the four storage questions for every connected toy or speaker in your home? If not, start with the one your youngest uses most.
Frequently Asked Questions
What did the Bondu AI toy leak expose?
About 50,000 chat transcripts between children and the toy, including names, birth dates, family details and intimate conversations, according to Malwarebytes. Anyone with a Gmail account could log in and read them. Bondu took the console offline within minutes of being notified.
Were children’s voice recordings leaked?
In the Bondu case, the reports describe text transcripts. In the Miko case, NBC News found the toy’s audio responses but no recordings of children’s own voices. Both still revealed personal details, because a toy’s replies often repeat what a child said.
How can I tell if my child’s AI toy is safe?
No toy is risk-free. Look for push-to-talk rather than always-on listening, a clear statement of what is stored and for how long, a way to delete history, and no third-party sharing. Common Sense Media recommends no AI toys at all for children 5 and under.
Can I delete what an AI toy has recorded?
Usually through the parent app or by contacting the company. In the U.S., COPPA gives parents the right to request deletion of data collected from children under 13. The FTC’s 2023 case against Amazon shows companies can be penalized for not honoring those requests.
Should I get rid of my child’s AI toy?
Not necessarily, but turn off cloud features you don’t need, delete stored conversations, and talk with your child about what they share with it. If the company can’t explain where data goes, retiring the toy’s AI features is a reasonable choice.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- Arntz, P. (2026, February 3). “An AI plush toy exposed thousands of private chats with children.” Malwarebytes. https://www.malwarebytes.com/blog/news/2026/02/an-ai-plush-toy-exposed-thousands-of-private-chats-with-children
- Perlo, J., & Collier, K. (2026, February 12). “AI toy maker exposed thousands of responses to children, senators say.” NBC News. https://www.nbcnews.com/tech/security/ai-toy-maker-exposed-thousands-responses-kids-senators-miko-rcna258326
- Federal Trade Commission. (2023, May 31). “FTC and DOJ Charge Amazon with Violating Children’s Privacy Law by Keeping Kids’ Alexa Voice Recordings Forever.” https://www.ftc.gov/news-events/news/press-releases/2023/05/ftc-doj-charge-amazon-violating-childrens-privacy-law-keeping-kids-alexa-voice-recordings-forever
- Common Sense Media. (2026, January 22). “AI Toys Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/ai-toys
- U.S. PIRG Education Fund. (2025). “AI comes to playtime: Artificial companions, real risks.” https://pirg.org/edfund/resources/ai-toys/
