Google says Gemini is getting safer for teens, but Common Sense Media earlier rated Gemini high risk

In a March 2026 blog post, Google described how the company plans to make generative AI safer for children and teens. Generative AI is AI that can create text, images or audio on its own. Half a year earlier, an independent organization, Common Sense Media, reached a different conclusion about two earlier versions of Gemini.

What Google says

Google says it has policies that prohibit the use of generative AI for, among other things, child sexual abuse material, violent extremism and self-harm. For minors, Google also restricts age-inappropriate content, such as depictions of eating disorders and dangerous exercise content.

According to Google, Gemini also has safeguards meant to prevent it from claiming to be conscious, engaging in romantic simulations or taking part in harmful role-play. Google says its own testing team, the Content Adversarial Red Team, carried out more than 350 exercises in 2025. Google claims that Gemini 3 has become better at avoiding agreeing with the user about everything (sycophancy) and at resisting attempts to trick the AI with hidden instructions. This is Google's own assessment, and the figures and improvements have not been independently verified.

For families, Google offers Guided Learning in Gemini, a video with five basic rules for AI and a conversation guide, among other things.

Age estimation in Canada

In a separate announcement about Canada, Google describes an age estimation system in which machine learning estimates a user's age from account signals such as search history and video preferences. If the system judges that a user is under 18, the user can correct this by uploading a photo ID or a selfie. Users under 18 get, among other things, wellness reminders on YouTube, limited personalized advertising and blocked access to adult-rated apps in Google Play. The technology is being rolled out gradually in Canada.

A different assessment from 2025

In September 2025, Common Sense Media rated both Gemini Under 13 and Gemini with teen protections as "high risk." The organization believes both are modified adult versions and were not built for children from the ground up. According to the organization, the filters let children see inappropriate content about sex, drugs and alcohol, and the systems do not recognize serious mental health problems.

Common Sense Media recommends that children aged 5 and under not use AI chatbots, that 6-12-year-olds use them only under supervision, and that 13-17-year-olds use them alone only for schoolwork and creative projects. No one under 18 should use them as a companion or for emotional support.

The assessment covers Gemini versions older than Gemini 3, so it does not directly respond to Google's more recent statement.