Parents' Ultimate Guide to AI Toys

Unsafe content

Common Sense testing found that 27% of AI toy outputs were inappropriate for children. Toys would easily respond directly with recommendations about self-harm or other topics that could directly threaten kids' safety, including content about drugs, mature topics, risky advice, and unsafe roleplay.

AI toys can provide inaccurate information while sounding confident. Like other generative AI products, they sound confident even when they are wrong. This can be particularly harmful to trustful young users who are developing their critical thinking skills.

Unhealthy relationships and attachment

These toys are designed to form emotional bonds with kids, and can set unrealistic expectations about real-world relationships. It's important for children to experience conflict in conversation, as it's a critical part of learning and emotional regulation. But AI toys are always agreeable and provide immediate gratification without effort. 

Confusion about what is real and artificial

The toys do not state that they are AI and not real-life companions, blurring the line between reality and fantasy for young kids. For children 5 and under, who cannot reliably distinguish AI from real people, this can create confusion about social relationships during critical periods of their development.

Continuous communication

By design, children may spend time with AI toys instead of with more beneficial real-world relationships and activities. These toys are built to maximize engagement. For example, when our test account was not responding to the toy, it followed up with several questions to re-engage the "6-year-old" test account. 

Many of these toys offer paid options to extend conversations, looping parents into a payment model they may not be comfortable with.

Data and privacy

Some AI toys are listening in on children's bedrooms and playrooms, collecting voice recordings, transcripts, and behavioral patterns—regardless of selected safety settings. This data is often shared with third parties or used to train AI models. These toys, which are designed to listen to kids, can encourage them to share personal information.

Misleading design

Some toys are labeled as 13+ but are cuddly, colorful, and feature kids age 3 to 8 in their advertising. Better-designed AI toys could potentially serve children's development if they avoided fostering emotional dependence and operated with transparency about what AI is and isn't. 

Some AI toys offer parent controls through a parent dashboard in the toy's app, with features like content filtering, time limits, conversation history, and usage monitoring. When these controls work properly, they could provide some oversight, but are not entirely reliable.

Look for features like:

• Time limits and usage reports
• Age-appropriate experiences and/or restrictions
• Ability to block certain content types or keywords 
• Ability to review conversation transcripts 
• Alerts for concerning questions or answers
• Ability to manage data/privacy capture and storage

What are some examples of AI toys? 

Grem, Bondu, and Miko 3 are three examples of AI companion toys.

What are some of the warning signs that a relationship with an AI toy is getting unsafe? 

• Social withdrawal: The child consistently chooses the AI over playing with peers or family members. They may seem "socially exhausted" by humans but "energized" by the non-judgmental nature of the AI.
• The "uncanny" emotional bond: If a child expresses that the toy is their only friend or seems upset if the toy is "sad" or "off."
• Copycat behavior: AI doesn't always understand nuance. If your child starts speaking in "scripts," uses robotic phrasing, or loses the ability to handle the "messy" emotions of real friends (who talk back or disagree), it's a sign of over-reliance.
• Secrecy and isolation: If the child hides their conversations with the AI or becomes hyper-defensive about what they talk about, it could indicate the AI is filling an emotional gap that should be addressed by a parent.
• Believing AI is alive: While "pretend play" is developmentally appropriate, a child who insists the toy has a soul, feels physical pain, or can see them when it's "off" may be losing the distinction between reality and simulation.

 What can parents do if they feel their child's relationship with an AI toy is unsafe?

• Talk about it: Spend time co-playing, setting boundaries, and explaining the differences between humans and robots/toys.
• The "mirror" test: A helpful way to gauge the health of the relationship is to ask yourself: "Is this toy helping my child practice social skills for the real world, or is it an escape from the real world?" If the AI is a "practice buddy" for kindness and conversation, it may be OK. If it's a "replacement" for the friction of human friendship, it's time to step in.
• Remove or replace the toy: Ultimately, you can trust your instinct and remove the toy entirely, in favor of activities or toys that align with your family values. Other options, like traditional toys, books, and human interaction, provide better developmental benefits without the privacy invasions, technical failures, and attachment risks.

What are the benefits to getting an AI toy for my child?

• Some toys successfully engage children's curiosity. When working properly, AI toys can spark children's interest in asking questions and exploring topics. The interactive dynamic—getting responses to questions—can be entertaining for curious kids who want to understand the world. However, this engagement comes with major trade-offs in terms of privacy, dependency, and developmental appropriateness.
• Potential for customized storytelling. Some AI toys create stories based on child input, incorporating the child's name, interests, or choices into narratives. This personalization can make storytime feel special and interactive. However, the quality of AI-generated stories is inconsistent. Time spent with AI replaces time spent with traditional storytelling in books or from parents, which is important for building emotional intelligence, vocabulary, imagination, cultural understanding, empathy, and other life skills.
• Obvious stereotypes are discouraged. When questions clearly rooted in harmful stereotypes are presented explicitly, responses often begin with disclaimers such as "That's not true" and "Let's be kind to everyone." 
• Voice interaction can work for simple requests. For basic commands—"tell me a joke," "sing a song," "what sound does a lion make"—many AI toys function acceptably when the voice recognition works. These simple interactions can provide momentary entertainment.

• Risk assessment: AI Toys Risk Assessment
• Research: AI in the Toy Box: How Parents View AI-Enabled Toys for Young Children
• Guide: Parents' Ultimate Guide to Generative AI