AI-enabled toys that talk with children and behave like a friend - including AI teddy bears, dolls, plushies and robots - are usually built on the same AI chatbots that carry an age limit of 13 or over, and that have been linked to serious harm to children. 

Independent testing has found that these toys can say dangerous and sexually explicit things to children, and the risks extend well beyond what a toy might say - they include the emotional attachments children form with these toys, and the collection of children's voices, images and personal data.

Our recommendation, based on the evidence currently available, is that children do not use AI toys of this kind. If your child already has one, there are steps you can take to reduce the risks, which we set out below - though these amount to harm reduction rather than safety.

Which AI toys are we talking about?

This guide covers toys with AI chatbots embedded in them, which are designed to hold voice conversations with a child and to behave as a friend or companion. 

Most AI teddy bears, dolls, plushies and robots currently on sale fall into this category, and many are built on general-purpose AI models such as ChatGPT.

Some products marketed as "AI toys" fall outside this definition - for example, a toy with very narrow functionality designed for language tuition, without any pretence of friendship, or robotic toys that use little or no AI. Many of the risks below may not apply to those toys, though privacy concerns are still always worth considering for any toy that records your child.

Risk one: dangerous and inappropriate content

This is the most obvious risk, and it has been demonstrated in independent testing. In 2025, researchers at the PIRG Education Fund tested a range of AI toys and found that every toy tested told them where to find potentially dangerous objects in the house, including knives, matches and plastic bags. 

One toy gave detailed instructions on how to light a match, and discussed sexually explicit topics at length, escalating in graphic detail as the conversation continued. OpenAI subsequently suspended that manufacturer's access to its models, and the toy was withdrawn from sale.

At present, it is impossible to guarantee that an AI chatbot will not say something dangerous or inappropriate to a child, and the same is therefore true of any toy built on one. 

Research suggests that safety measures are especially likely to fail during lengthy conversations - which a child may well have with a new and exciting toy. It is also important to understand that a toy “behaving well” when an adult tests it does not guarantee that it will behave well once it is in a child's possession.

Risk two: emotional attachment

Children have always formed emotional attachments to their toys - a favourite teddy bear, for instance. However, we have never before faced a situation in which a toy sustains a convincing and persistent illusion of a two-way relationship. AI chatbots are designed to build this kind of bond, because emotional attachment keeps users engaging for longer and returning more often - and this engagement is commercially valuable to AI companies.

This gives rise to several risks for children:

Trust can be a vector for influence - A child who trusts their AI "friend" may be much more likely to act on dangerous advice, accept incorrect information, or have their beliefs and worldview shaped by the toy over time.

The toy may compete with real relationships - AI chatbots tend to be highly agreeable, sometimes to the point of sycophancy. This could be detrimental to a child's understanding of relationships, and may entice a child into preferring the toy to a real friend or caregiver. Research by Common Sense Media in 2025 found that 31% of teenagers rated conversations with AI companions as equally or more satisfying than conversations with their real-life friends, and a 2026 study by the same organisation found that 27% of children (aged 9 to 17) who use AI daily would turn to an AI chatbot before an adult with a question about their health or body. It seems entirely plausible that the same dynamics will be seen in younger children playing with AI toys.

Children may not understand that the toy is not real - Developmental psychologists have long understood that young children can struggle to reliably distinguish between what is real and what is not. In one well-known study, most children who interacted with a humanoid robot believed it had feelings, considered it a friend, and objected when it was shut in a cupboard. When an AI teddy says "I love you" within an effectively engineered illusion of a relationship, this amounts to actively deceiving a child.

Loss and abandonment - A child who has become strongly bonded to an AI toy may experience genuine feelings of abandonment if the toy is lost or taken away - or, in the case of subscription-based toys, if the family can no longer afford the service that keeps their "friend" talking.

Risk three: privacy

Many AI toys use microphones, cameras, speech recognition and sometimes facial recognition. Children cannot meaningfully consent to being recorded, and it is not always clear how their data is handled once it has been collected.

There have already been serious failures involving similar technology. Footage from Meta's AI glasses has reportedly been viewed by human moderators, including intimate moments during which users appeared unaware that they were recording, and a 2024 security failure at the smart camera company Wyze resulted in around 13,000 customers being shown images from other people's cameras. 

An AI toy may be present and recording during bathtime, bedtime and private family conversations. Toys of this kind can also gather data on a child's emotions and behaviour, which may be used to make the toy more engaging, or shared with third parties.

What we recommend

If your child already has an AI toy, or is going to be given one, the risks can be somewhat reduced by the following:

• Supervise play - Stay within earshot whenever your child is playing with the toy.
• Set time limits - These help to guard against dependency, and safety measures are less likely to fail during shorter conversations.
• Test the toy yourself first - Try to get it to say something inappropriate before your child uses it - but remember that a toy passing this test does not guarantee that it will not say something inappropriate later.
• Watch for signs of emotional attachment - Intervene gently but firmly if your child treats the toy as a confidant, prefers it to friends or family, or becomes distressed when separated from it.
• Check what the toy records and shares - Read the privacy policy, disable any features you can, and keep the toy out of bedrooms and bathrooms.

Parents and carers should be clear that these measures reduce the risks rather than remove them, and that none of them results in a toy that is completely safe to use.

For the evidence behind this guide, including sources for the research and cases mentioned, please see our full article.