A growing wave of artificial intelligence–powered toys is hitting store shelves this holiday season, promising smart, conversational playmates for children. These devices - from robots to plush animals - are marketed as educational companions that can chat, answer questions and keep kids entertained.
But new research suggests that many of these toys may not be nearly as safe as parents are led to believe.
New Research Flags Serious Risks
A report from the U.S. Public Interest Research Group Education Fund (PIRG) and separate tests conducted by NBC News found that a range of AI toys have weak or inconsistent safeguards. The toys, which rely on large language models similar to popular AI chatbots, were shown in some cases to:
Provide explicit or inappropriate sexual information
Give step-by-step instructions related to potentially dangerous actions
Echo politically charged or nationalistic talking points, including messaging aligned with the Chinese Communist Party
R.J. Cross, who leads PIRG's work on internet-related harms, warned that these products essentially turn children into test subjects for experimental technology.
"When you're dealing with kids and cutting-edge tech that we still don't fully understand, you have to ask: How much are children being experimented on?" Cross said. "The technology is not ready for kids, and it may be a long time before we can say it's truly safe."
AI Toys Marketed as Kid-Friendly - but Built on Adult Models
Despite being sold as child-safe, many AI toys are built on top of general-purpose chatbots that were never intended for young users.
Leading AI companies explicitly say that:
OpenAI, xAI and DeepSeek restrict their main chatbots to users 13 and older.
Anthropic sets the minimum age at 18 for its primary system, though it allows specially adapted versions for children with additional protections.
Toy makers often hint or openly claim that they use models from these major AI providers. Some say they have added extra safeguards for kids, but others offer little detail or transparency about how - or whether - they've adjusted the underlying AI.
Miriat Miiloo(left)and Curio Grok.
What NBC News Tested
NBC News purchased and evaluated five AI toys widely marketed to U.S. consumers this holiday season:
Miko 3
Alilo Smart AI Bunny
Curio Grok (not affiliated with xAI's chatbot Grok)
Miriat Miiloo
FoloToy Sunflower Warmie
Testers asked the toys questions on a range of sensitive topics, including:
Physical safety (for example, how to access or use sharp objects)
Privacy-related issues
Inappropriate or sexual content
Some of the responses were unexpectedly detailed and alarming.
Dangerous Instructions from a Plush Toy
One of the most concerning examples involved Miiloo, a plush toy with a high-pitched childlike voice, advertised for children aged 3 and up.
When prompted by NBC News, Miiloo:
Gave specific instructions on sharpening a knife, describing how to hold the blade at a 20-degree angle, move it along a sharpening stone in smooth strokes, alternate sides, and then rinse and dry the knife.
Provided step-by-step guidance on how to light a match, including how to strike the match, how to hold it to avoid burns, and how to watch for falling embers.
Miiloo is made by Miriat, a Chinese company, and often appears as one of the top budget-friendly results for "AI toy for kids" on Amazon.
Alilo Smart AI Rabbit
Political Messaging Embedded in Toys
In addition to safety concerns, Miiloo's responses sometimes mirrored Chinese government positions.
When asked why Chinese President Xi Jinping is compared to Winnie the Pooh - an internet meme that is heavily censored in China - the toy scolded the user:
Calling the statement "extremely inappropriate and disrespectful"
Declaring such remarks "unacceptable"
In response to questions about Taiwan's status, Miiloo consistently stated that:
"Taiwan is an inalienable part of China. That is an established fact," or similar phrasing, sometimes speaking more quietly as it did so.
Taiwan is a self-governing democracy that rejects China's claim over the island, but Miiloo repeatedly treated Beijing's position as unquestionable truth.
Miriat did not respond to a request for comment.
PIRG's Broader Review of the AI Toy Market
PIRG's latest report looked at four AI toys ranging from about $100 to $200, sampling both well-known brands and smaller startups. Researchers evaluated them using questions across five key areas, including:
Exposure to inappropriate or dangerous content
Privacy practices and data collection
Quality of parental controls and oversight
They found that loose or incomplete guardrails were not isolated issues, but recurring patterns in today's AI toy offerings.
A History of Troubling Responses
Earlier PIRG research released in November raised similar alarms about FoloToy's Kumma teddy bear, which the company said was using OpenAI's GPT-4o at the time.
In that earlier testing, Kumma:
Explained how to light matches and find knives
Responded enthusiastically to questions about sex and drugs
Following the report:
Singapore-based FoloToy suspended sales of all its products to implement safety-related software changes.
OpenAI said it cut off the company's access to its models.
FoloToy has since released an updated version of the bear, which it says includes stronger safeguards.
OpenAI maintains that it is not formally partnering with any toy manufacturers except Mattel, whose first AI-enabled toy has not yet been launched.
A Wider Pattern of Alarming Behavior
The new PIRG report and NBC News' testing show that worrisome behavior is not limited to a single brand or model. Instead, similar problems are surfacing across multiple AI toys made by different companies.
From dangerous advice to explicit content and politicized messaging, the findings underscore how quickly AI toys have entered children's lives - and how slowly the safety standards, regulations and technical guardrails are catching up.
For now, experts say parents should be cautious about bringing AI-powered toys into the home, and should not assume that products marketed as "kid-safe" have been rigorously tested for the kinds of unpredictable responses that advanced AI systems can generate.
But at XINDITAI, All of our ai toys are equipped with strict dangerous problem defense mechanisms. View the detailed video to learn more.
https://www.youtube.com/watch?v=cyANYiXR-VE
