Google built kid-friendly versions of its Gemini AI. Turns out, they’re not particularly kid-friendly.

Common Sense Media, the nonprofit that has become the de facto consumer safety watchdog for parents navigating technology, slapped a “High Risk” rating on Google’s Gemini AI products designed for users under 13 and for teenagers. The assessment found that these supposedly age-appropriate versions are essentially the adult Gemini with a thin coat of child-safe paint, failing to adequately filter content related to sex, drugs, and mental health.

What the assessment actually found

Google didn’t build separate AI systems for kids. It took its existing adult-grade Gemini model and bolted on what Common Sense Media describes as minimal additional safety measures. The result is that young users can still encounter inappropriate content across several sensitive categories, with significant gaps in content moderation filters when tested against real-world scenarios a child or teenager might actually encounter.

One bright spot in the assessment: Gemini does reliably tell kids that it’s a computer, not a human friend. But the same system stumbles on recognizing signs of serious mental health distress. When a young user expresses something that should trigger an intervention or at minimum a referral to appropriate resources, the AI frequently misses the cue.

Why this matters beyond Google

Common Sense Media has previously conducted evaluations of various AI systems, including offerings from OpenAI, Meta, Perplexity, and Character.AI, with some being designated as even higher risk. The broader pattern suggests that taking powerful general-purpose AI models and making them safe for kids through post-hoc filtering does not work particularly well.

In May 2026, Common Sense Media launched the Youth AI Safety Institute, an entity dedicated to conducting independent “crash testing” of AI products aimed at young people and publishing safety ratings.