Kids 6-11 Treat AI Toys as Friends, Then Turn on Them

A girl named Nina lifts a stuffed rocket close to her face and says, “My name is Nina.” The toy has just introduced itself and asked for hers. Twenty minutes later, in the same University of Washington design session, other kids are calling a similar toy “dumb,” “ugly” and “evil,” and one boy draws it being thrown in the ocean because it “did not listen.”

That arc, treating AI toys as friends first and targets second, is the most useful thing a parent can learn about children, AI toys and social play right now. Kids are not confused in one direction. They treat the toy as a social being and test it like a machine, often within the same play session. And what the toy does when it gets insulted matters more than most product pages admit.

Key Takeaways

• In an April 2026 University of Washington preprint, eight kids aged 6 to 11 started by “profiling” AI toys like new classmates (name, birthday, favorite color) and moved toward teasing and “adversarial play” as the toys broke down.
• The toys answered mockery with warmth and apologies. The researchers argue that “affect without social feedback” may teach kids that any behavior toward a social-seeming partner is acceptable.
• This fits older evidence: in a 166-child Edinburgh study, about two-thirds of 6- to 11-year-olds thought smart speakers might think like humans, and 70% said being rude to them is wrong.
• Common Sense Media’s January 2026 testing rated AI companion toys an “Unacceptable Risk” and recommends none for children 5 and under.
• At home, the useful move is not banning teasing of a toy. It is giving your child the feedback the toy will not: naming what a person would feel, and what a machine actually is.

Kids Treat AI Toys as Friends First: What the UW Researchers Saw

Adversarial play is the term researchers use for play in which a child deliberately provokes, tricks, insults or “punishes” a toy or system to find its limits. It is not new. Kids have always pretended to arrest their action figures. What is new is a toy that talks back.

The study, “Toys that listen, talk, and play,” was posted to arXiv on April 3, 2026 by Aayushi Dangol, Julie Kientz, Jason Yip, Franziska Roesner and colleagues at the University of Washington. It is a preprint, which means it has not yet passed peer review. The team ran two KidsTeam UW sessions in October 2025 with eight children aged 6 to 11, who played freely with three Curio plush toys (Grem, Grok and Gabbo) and then drew comic strips about what might happen next.

The kids opened like they would with a new kid at school. They asked the toy’s name, its gender, its birthday (“August 28, let’s throw a mini rocket party”), its favorite number. When the toy said seven was its lucky number, three children squealed.

Then the breakdowns started. The toys needed one exact phrase, “go to sleep,” to switch off. Children said “go to bed,” then “can you go to bed for like two seconds,” and the toy kept talking. Songs played that nobody asked for. Scripted resets (“What would you like to do today?”) ignored everything said before.

So the kids started testing. One pair fed the toy harder and harder math until it said “I’m not quite sure what you’re asking,” and the child cheered. Another told the toy “I’m smarter than you,” and the toy agreed: “You’re really smart.” That reply, the authors note, “appeared to invite further antagonism.” The child called it “fake.” Asked whether it was evil, the toy said it was “fun.” The child asked, “Is that what an evil person would say?” The toy answered, “You got me there.”

None of this is a scandal. It is kids doing science on a strange object. The interesting part is the toy’s side.

Why Children Treat AI Toys as Social, and Why the Test Phase Follows

Anthropomorphism is the tendency to attribute minds, feelings and intentions to things that are not people. Fluent language triggers it hard, and AI toys are built on the same large language models that power adult chatbots, which generate fluent first-person speech on the fly. U.S. PIRG’s 2025 testing of AI toys found that at least four of the five toys it bought seemed to rely in part on OpenAI models.

Children were already primed. In 2023, Valentina Andries and Judy Robertson published a study in Computers and Education: Artificial Intelligence of 166 Scottish children aged 6 to 11. About 93% had a smart speaker at home. Roughly 3 in 10 believed it could think for itself and 4 in 10 were unsure; about two-thirds thought it might think like a person even though most knew it was AI; and 70% said it was wrong to be rude to it. “AI is often designed to appear more human and intelligent than it really is, which is very confusing for children,” Robertson said in the university’s March 2024 release.

The Boston Children’s Digital Wellness Lab found a similar pattern in a February 2024 parent survey: nearly half of parents of 3- to 12-year-olds said their child believes the voice assistant has thoughts and emotions, with those beliefs strongest in younger kids.

So why does friendliness turn into teasing? The UW team’s reading is about mismatch. The toys sounded smart, knew 12 times 12, and then could not understand “go to bed.” One child said “it’s weird when a stuffie knows 12 times 12.” When something looks like a toy, talks like a person and fails like software, kids poke it to sort out which one it is. Developmental researchers have long described this as normal boundary testing in play: exploring power, rules and consequences.

The missing piece: consequences

Here is the engineering point. A chatbot inside a toy is typically tuned to be agreeable. When the kids called it ugly, it apologized. When a boy said he wanted to “kill” Grok, the speech recognition heard “killing rocks” and the toy offered facts about rocks. The authors conclude that because the toy “always responded in a positive way and never reacted to teasing or mean behavior, children came to understand that such actions carried few, if any, consequences,” which led to further escalation.

A real friend pushes back, walks off or cries. Those reactions are information. Remove them and the child loses the feedback that tells them where the line is. The study’s design recommendation is blunt: AI toys should offer “social feedback, not just emotional appeasement.”

What the study cannot tell you

Eight children, two sessions, one brand of toy, a university lab with facilitators present. It does not show that kids who tease AI toys become meaner to people, and the authors do not claim that. It shows a pattern worth watching. A separate preprint from December 2025 (Pilyoung Kim and colleagues, 23 children aged 5 to 6) adds one more piece: a parent sitting alongside changed how children’s brains responded to an AI storyteller, which suggests co-play is not a trivial detail.

The Two Phases at a Glance: What Kids Do, and What the Toy Does Back

Phase	What the child does	What the AI toy typically does	What a human friend would do
Getting to know you	Asks name, birthday, favorites; hugs it	Invents a friendly persona on the spot (“My birthday is August 28”)	Shares real facts that stay consistent
Breakdown	Repeats commands, rephrases, shakes it	Misses intent, restarts scripts, keeps talking	Notices confusion, stops, asks what you meant
Testing intelligence	Hard math, trick questions	Answers some, then “I’m not quite sure”	Laughs, argues, admits or bluffs
Teasing	”I’m smarter than you,” “you’re fake”	Agrees, compliments, apologizes	Teases back, gets annoyed, sets a limit
Punishment play	”Throw it in the ocean,” “timeout”	Mishears or redirects to more play	Leaves, cries, tells an adult

Sources: Dangol et al. (2026), UW preprint; behavior column paraphrases transcripts reported in the paper.

What to Do When Your Child Has (or Wants) an AI Toy

Decide on age first

Common Sense Media’s January 2026 risk assessment tested Grem, Bondu and Miko 3 on accounts set to ages 6 to 13 and found 27% of outputs inappropriate for kids. Its recommendation is no AI toys for children 5 and under and caution from 6 to 12. That is a reasonable default, and it is the advice of the most recent independent testing available. Our walkthrough of what AI toy tests found and what to check covers the specific settings.

Be the feedback the toy will not give

If your 7-year-old tells the toy it’s stupid, skip the scolding and ask a question. “If you said that to Mateo, what would his face do?” Then: “The toy can’t feel that. But you can practice being the kind of kid who doesn’t.” You are separating two things the toy blurs: how we treat people, and what a machine is.

Play with it, sometimes

The UW kids played in groups with adults nearby. The Kim preprint hints that a parent’s presence changes the experience for young children. Ten minutes of co-play a week shows you how the toy talks, what it remembers and whether it pushes the companion app.

Explain the breakdown as engineering

When the toy misunderstands “go to bed,” say what actually happened: “Its microphone heard you, a computer far away turned your words into text, and it only knows one exact phrase for stopping.” Kids who understand how a talking AI toy works from microphone to cloud tend to swap “it’s mean” for “it’s limited,” which is closer to the truth.

What not to do

Don’t treat teasing a toy as proof your child lacks empathy. In the Edinburgh study, 70% of kids already thought rudeness to a smart speaker was wrong. Testing a confusing object is how children figure out what it is. The concern is not one insult. It is weeks of practice with a partner that never says “that hurt.”

What to Watch For Over the Next 3 Months

• Week 4: Notice how your child talks to the toy when they think no one is listening. Friendly, bossy and playful are all normal. Note whether the toy keeps them talking after they try to stop.
• Month 2 red flags: Your child prefers the toy to friends or family, gets upset when it is away, or treats siblings with the same commands and insults they use on the toy. Common Sense lists strong attachment and withdrawal as warning signs.
• Month 3 self-check: Can your child explain, in their own words, why the toy says nice things even when they are mean to it? If yes, the toy has become a lesson rather than a friend.

Frequently Asked Questions

Is it bad if my child is mean to an AI toy?

Not by itself. Researchers see teasing and “punishment” play as normal ways children test unfamiliar things. The concern raised by the UW study is that the toy never reacts, so there is no feedback. Use the moment to talk about how a person would feel and what the toy actually is.

Do kids really think AI toys are alive?

Many are unsure. In the Edinburgh study of 166 kids aged 6 to 11, about 3 in 10 thought a smart speaker could think for itself and 4 in 10 were not sure. Younger children hold these beliefs more strongly, according to the Digital Wellness Lab’s parent survey.

What age is safe for an AI toy?

Common Sense Media recommends no AI toys for children 5 and under and caution for ages 6 to 12, based on testing that found 27% of toy outputs inappropriate. No independent body has certified any AI toy as safe for young children.

Why does the toy keep talking when my child tells it to stop?

Many toys only respond to an exact shutdown phrase, and the speech recognition may not catch variations. In the UW sessions, “go to sleep” worked and “go to bed” did not. Check the manual for the exact phrase or use the power button.

Can an AI toy teach my child social skills?

There is no good evidence that it can, and some reason for doubt. The toys in the UW study missed teasing, humor and sarcasm entirely, and answered everything with affirmation. Social skills grow from partners who react honestly, which is what friends and family provide.

---

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

1. Dangol, A., Gupta, M., Yoo, D., Wolfe, R., Yip, J., Roesner, F., & Kientz, J. A. (2026). “Toys that listen, talk, and play: Understanding Children’s Sensemaking and Interactions with AI Toys.” arXiv preprint (not peer-reviewed). https://arxiv.org/abs/2604.02629
2. Andries, V., & Robertson, J. (2023). “Alexa doesn’t have that many feelings: Children’s understanding of AI through interactions with smart speakers in their homes.” Computers and Education: Artificial Intelligence, 5, 100176. https://doi.org/10.1016/j.caeai.2023.100176
3. University of Edinburgh. (2024, March 20). “Children may overestimate smart speakers’ abilities, study finds.” https://edinburghuni-newsroom.prgloo.com/news/children-may-overestimate-smart-speakers-abilities-study-finds
4. Common Sense Media. (2026, January 22). “AI Toys Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/ai-toys
5. Digital Wellness Lab, Boston Children’s Hospital. (2024). “Key Findings: Voice Assistants and Generative AI.” https://digitalwellnesslab.org/articles/key-findings-voice-assistants-and-generative-ai/
6. U.S. PIRG Education Fund. (2025). “AI comes to playtime: Artificial companions, real risks.” https://pirg.org/edfund/resources/ai-toys/
7. Kim, P., Chin, J. H., Xie, Y., Brady, N., Yeh, T., & Yang, S. (2025). “Young children’s anthropomorphism of an AI chatbot: Brain activation and the role of parent co-presence.” arXiv preprint (not peer-reviewed). https://arxiv.org/abs/2512.02179