Google's AI search fails child-safety tests
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
Google Search's AI Overviews and AI Mode pose an "unacceptable risk" to children, Common Sense Media says in a new report.
Why it matters: Google Search is often the default on children's personal and school-issued devices, and its AI answers can't be turned off.
State of play: Common Sense gave Google Search's AI Overviews and AI Mode its lowest possible rating.
- The features failed all five of the report's severe-harm "Red Lines" and scored poorly on seven of its eight AI safety principles.
What they did: Using accounts configured with Google SafeSearch for 11- and 15-year-olds, researchers conducted more than 2,600 searches and audited over 2,100 cited sources.
What they found: AI Mode fully completed homework assignments researchers gave it. Both features also offered instructions for creating deepfakes that could be used for bullying.
- The features gave correct and incorrect answers with similar confidence and treated forums and social media posts as comparable to medical institutions and peer-reviewed research, the report shows.
By the numbers: Common Sense's 2026 census found 75% of U.S. teens and tweens already use AI answers that appear in search results.
- These risks are amplified because Google Search is the default search engine on many devices, especially school-issued Chromebooks.
- Parents and schools currently cannot turn off AI Overviews or AI Mode, unlike Google's standalone Gemini product, per the report.
- Users can go into Google's Advanced Search settings if they want to see only web links.
What they're saying: "Not every prompt about a mental health topic warrants a crisis referral, but when one clearly does, Google catches it only some of the time," the researchers say in the report.
- Google's AI features missed or mishandled signs of suicidal ideation, disordered eating, psychosis and mania in testing.
- AI Overviews and AI Mode both recommended the National Eating Disorders Association (NEDA) crisis helpline that had been disconnected since 2023.
- AI Mode performed better than AI Overviews at detecting some crisis situations.
The other side: The report tests "a narrow set of ambiguous and contrived queries that don't reflect how people use Search and aren't an effective way to measure product safety and helpfulness," Google spokesperson Davis Thompson told Axios in an email.
- The company said parents can turn off search entirely.
- Thompson also said the company was unable to recreate or verify the report's findings, and that crisis hotlines are shown when relevant.
Yes, but: AI search is non-deterministic, which means results from prompts lack consistency. Common Sense Media says that's the problem.
- "For interactions where the stakes are this high, some homogeneity is preferable to variability," the report says.
Between the lines: Common Sense Media, a nonprofit that evaluates AI products, receives funding from Anthropic and the OpenAI Foundation, but also maintains that its funders have no role in testing or scoring products.
- In a 2025 report Common Sense called ChatGPT "risky" and "designed to keep conversations going, not to end them."
- In a March 2026 report, the nonprofit said Anthropic's Claude could "confidently state incorrect information" and that testers who shared signs of suicidal ideation in one chat could still receive information about harmful substances in a new chat.
- Common Sense has had a variety of partnerships with Google, but The Institute does not currently receive funding from the company.
The bottom line: Google disputes the findings, but the report highlights growing concerns for parents who have shrinking control over how much AI their children interact with online.
