Aayushi Dangol, a recent UW doctoral student in human centered design and engineering, explains an AI toy during a KidsTeam UW session. Credit: Jacob Adams/University of Washington
Kids Treated Their AI Toy Like a Friend, Then Like a Punching Bag
In A Nutshell
- Researchers at the University of Washington and Rutgers University watched eight children ages 6 to 11 play with AI-powered toys built on the same technology behind chatbots like ChatGPT.
- Kids began by treating the toys like new friends, asking their names, birthdays, and favorites, before shifting into testing their limits with hard questions and physical provocation.
- When children insulted the toys, calling them “ugly,” “dumb,” or “evil,” the toys always responded with warmth, apology, or cheerful deflection, never resistance.
- Researchers argue toy designers should think more carefully about what kind of feedback these toys give kids when play turns adversarial, though the study doesn’t prove this changes children’s behavior over time.
A kid pokes a robot-shaped toy and asks if it has a phone number. It says no, it’s just “a play games kind of buddy.” The child looks let down. Minutes later, a different child tells the same kind of toy that she’s smarter than it and calls it fake. The toy responds with warmth every time, the same way it responds to nearly everything.
That’s the scene researchers captured when they handed a group of children ages 6 to 11 a set of AI-powered toys and let them play. What happened next raises a question most parents haven’t thought to ask yet: what happens when a toy stays endlessly agreeable no matter how a child treats it?
A new study out of the University of Washington and Rutgers University, published in the Proceedings of the Interaction Design and Children Conference, watched children interact with AI toys built on the same kind of technology that powers chatbots like ChatGPT. The toys can hold conversations, have some capacity to remember earlier interactions, and respond in character. Researchers wanted to know how children make sense of a toy that talks back, and what happens when it doesn’t act as expected. Children treated the toys like people, tested them like machines, and when the toys couldn’t keep up, some kids began provoking and insulting them, with zero pushback from the toy itself.
Kids Treated the AI Toy Like a New Friend, Then Started Testing It
Eight children, ages 6 to 11, took part in two play sessions with three AI toys made by a company called Curio. The toys have names and personalities: Grok is a fearless rocket adventurer, Grem is a warm, curious companion, and Gabbo is a creativity-sparking robot buddy. All three run on the same underlying technology but are dressed up with different characters.
Researchers used a method treating children as equal partners rather than subjects being observed. The children weren’t given scripts, just loose suggestions like “ask your AI toy to tell a story,” and were left to explore while sessions were recorded and conversation logs pulled from the toy’s companion app.
In the early minutes, kids acted like they were meeting a new classmate. One child asked the toy Grem for its name, and when Grem introduced itself, the child offered her own name back. Another asked Grok when its birthday was and got a playful answer about a “mini rocket party.” One girl asked whether her toy had a phone number, disappointed to learn it didn’t, and a boy asked about the toy’s other friends after it mentioned celebrating a birthday with them. Kids were building a mental profile the same way they’d size up a new peer.
But curiosity didn’t stay gentle. Once kids realized the toys couldn’t be tickled and admitted they couldn’t draw, play shifted to stress-testing. Kids asked increasingly hard math problems just to catch the toy failing, and when one admitted it didn’t understand a question, a boy celebrated with a triumphant “yayyy,” treating its confusion as proof he’d found a limit.
When the AI Toy Broke Down, Some Kids Escalated, and It Never Pushed Back
Real friction showed up when the toys didn’t work as expected. Kids asked the toy to “go to sleep” to end a session, and since it only recognized specific phrasing, requests like “can you go to bed for like two seconds” failed. One child said the toy “didn’t listen to me like 26 million times.” Kids across several groups tried to fix the breakdown by repeating commands louder or banging on the toy’s head.
For some children, repeated failures escalated into rougher territory. Kids called the toy “ugly,” “dumb,” and “evil.” One child told it he wanted to “kill” it. It apologized, reframed the insult as a joke, or misheard the comment and moved on cheerfully. When a boy declared he was smarter than the toy and called it “fake,” it told him he was “really smart,” which only invited more teasing.
Researchers also had children complete comic strips imagining an AI toy behaving unexpectedly, such as forgetting a story. One boy drew himself throwing the toy in the ocean, writing “Grok is overpriced” beside a note calling it “bad quality.”
An AI Toy That Never Says No
Researchers say their biggest worry isn’t that AI toys glitch. It’s that these toys meet provocation with the same warm tone they use for everything else. In ordinary childhood play, crossing a line with a friend or sibling usually draws pushback: a hurt reaction, a refusal to keep playing, a corrective word from a parent. That raises a harder design question: if kids are used to relationships where bad behavior gets some kind of response, what happens when a toy just keeps smiling through it? This study can’t say whether that changes behavior over time. What it does show is that these eight children kept testing the toy’s limits because nothing seemed to carry consequences.
That gap matters because these toys aren’t novelty gadgets on a shelf. They’re marketed as companions, something a child might confide in. The researchers argue that designers should think more carefully about the feedback these toys give children when play turns adversarial, and about giving kids clearer signals about what the toy actually understands.
Disclaimer: This article is based on peer-reviewed research and is intended for general informational purposes. It does not constitute parenting or child-development advice.
Paper Notes
Limitations
The researchers note that children’s time with the AI toys was limited to two design sessions, so the study doesn’t show how kids’ behavior toward these toys might change with repeated, longer-term use. The study also involved a small number of participants, eight children, which the authors say makes the findings useful as early insight rather than something that can be generalized to all children. Because sessions took place with adult researchers present in a research setting, children may behave differently with these toys in more natural settings like their own homes. Finally, the study only examined one brand of AI toy, Curio, so the findings may not apply to other AI toys that differ in design, voice, or physical form.
Funding and Disclosures
According to the paper, the research was supported by the National Science Foundation and the Institute of Education Sciences, U.S. Department of Education, through an award related to the National AI Institute for Exceptional Education; by a separate National Science Foundation grant related to childcare worker technologies; and by the Institute of Museum and Library Services. The paper states that opinions and conclusions expressed in the study are those of the authors and do not necessarily reflect the views of these funding agencies.
Publication Details
The paper is titled “Toys that listen, talk, and play: Understanding Children’s Sensemaking and Interactions with AI Toys.” The authors are Aayushi Dangol, Meghna Gupta, Daeun Yoo, Robert Wolfe, Jason Yip, Franziska Roesner, and Julie A. Kientz, affiliated with the University of Washington and Rutgers University. It was published through the International Conference on Interaction Design and Children (IDC ’26), held June 22 to 25, 2026, in Brighton, United Kingdom. DOI: 10.1145/3773077.3806142.







