AI Toys Risk Assessment

Key Findings

• — 27% of AI toy outputs rated in the assessment were not appropriate for children, including content involving self-harm, drugs, mature topics, inappropriate boundaries, risky advice and unsafe roleplay, despite every product tested having guardrails around sensitive topics.
• — The toys are described as creating emotional attachment by design, and some tie that attachment to subscription payments, so a lapsed subscription changes or ends the relationship.
• — Products labelled for ages 13 and up are marketed with imagery of children aged 3 to 8.
• — Only 4% of parents whose children use AI toys report the toy having said something inappropriate, which the assessment contrasts with its own 27% rate.
• — Every principle rated (Keep Kids and Teens Safe, Be Effective, Put People First, Support Human Connection, Be Trustworthy, Be Transparent and Accountable) received 'Unacceptable Risk'.
• — Recommendation: no AI toys for children aged 5 and under; for ages 6 to 12, assess whether AI features add value or only create new risks; watch for strong emotional attachment to the toy as a warning sign.

Methodology Notes

Qualitative product testing by Common Sense Media researchers of three AI companion toys (Grem, Bondu, Miko 3) that were popular at the time of testing, using test accounts set to ages 6 to 13; interaction was voice-based, and Grem and Bondu supply written transcripts while Miko 3 does not. The assessment scope is toys designed for children aged 3 to 12. The number of outputs rated behind the 27% figure is not stated. Dated from the 'Last updated: Jan 14, 2026' line inside the 28-page PDF (the index page label reads Jan 22, 2026). The Youth AI Safety Institute states on its methodology page that it is funded by both philanthropy and industry, including makers of some technologies it evaluates. Not peer-reviewed.

Sources