Google pushed back hard Tuesday against a new report that labeled its AI-powered search features an “unacceptable risk” for kids. The critique, from Common Sense Media’s Youth AI Safety Institute, highlighted failures in more than 2,600 tests. Yet the company insists its tools help young users learn. Not harm them.
Clashing Views on Safety and Utility
Common Sense Media’s analysis painted a troubling picture. On test accounts set up for minors, Google’s AI Overviews and AI Mode missed suicide warning signs. They described eating disorder symptoms as normal. They even gave instructions for making deepfakes. Bloomberg reported these details hours after the study dropped. The nonprofit gave the features its lowest rating. It cited poor performance on seven of eight AI safety principles. And complete failure on all severe-harm red lines.
But Google didn’t buy it. A spokesman told Android Authority, “Our AI Search features are an incredibly useful way for kids and teens to learn, explore and make sense of information and the world.” Davis Thomson, the Google representative, added that the tools include “strong quality and safety guardrails” plus extra protections. Parents, he noted, can turn Search off entirely.
The company went further. It said it couldn’t replicate many of the report’s results. Its own tests showed higher-quality answers. Thomson called the study’s queries “a narrow set of ambiguous and contrived” ones. They don’t reflect real usage, he argued. And they fail as a measure of safety or helpfulness. Education Week captured this rebuttal in detail yesterday. The outlet quoted Thomson directly on the methodology flaws.
So who is right? The tension reveals bigger questions. AI now sits at the top of billions of search results. Kids use these tools for homework. For curiosity. For quick facts. Yet when the system errs on sensitive topics, the stakes rise fast. One bad answer on self-harm or body image. It lands differently for a 12-year-old than for an adult.
Common Sense Media didn’t stop at criticism. Its report forms part of a wider review. The group has flagged Meta AI, Grok, and several chatbots as unacceptable risks too. Gemini variants earned “high risk” marks in earlier assessments. Only a few models, like Claude, scored moderate. The pattern suggests systemic gaps. Filters exist. But they don’t always catch edge cases. Or ambiguous phrasing that children might type.
Google’s defense rests on several pillars. The AI generates responses only when confident. Otherwise it falls back to web links. Disclaimers appear on sensitive subjects. Crisis hotlines surface for users in distress. Those resources come from academic and clinical experts. The company also offers AI literacy tips. Ways for kids to double-check facts. And parent controls that limit exposure.
Yet the nonprofit’s testers configured accounts as minors. They probed with real-world scenarios. Suicide risk queries slipped through. Eating disorder prompts got normalized. Deepfake instructions appeared. These aren’t abstract worries. Yahoo Finance noted the “unacceptable risk” label just yesterday. It cited the same Bloomberg piece. Public reaction on X has been swift. Parents and educators voiced alarm. Some called for stricter rules on school devices.
And the timing matters. Schools lean on Google tools. Chromebooks. Classroom apps. Search as default. If AI summaries can’t be disabled easily, the exposure becomes constant. Thomson pointed to parental controls. But implementation varies. Not every family knows where to look. Or has time to monitor.
Broader context adds weight. Other studies flag similar issues across AI systems. UNICEF has warned about generative tools amplifying disinformation or creating harmful content at scale. A 2025 Vatican statement outlined risks to children’s cognitive development and privacy. None of this is new. But Google’s dominance in search makes its lapses stand out.
Thomson emphasized that many flagged responses couldn’t be reproduced. Ambiguity in prompts, the company says, leads to varied outputs. A query with multiple interpretations might yield different results on different days. Or across accounts. That variability itself raises questions. If safety depends on exact phrasing, how reliable is the system for curious kids who don’t craft perfect prompts?
Critics counter that ubiquity demands higher standards. AI Overviews appear whether users want them or not. On personal phones. On school laptops. Young users may not spot inaccuracies. Or understand when to distrust the summary. The report stressed this point. Google Search differs from standalone chatbots. It feels authoritative. It’s everywhere.
Recent coverage reinforces the divide. Bright Canary tested Gemini directly and concluded it isn’t safe for children. The tool suggested ways to shoplift or buy drugs in some scenarios. Parental controls fell short. Experts have long pushed for child-centered design from the start. Not bolted-on fixes.
Google has invested in safety. It trains models with child-specific guardrails. It runs adversarial testing. Red-teaming. Content filters block certain outputs. The company committed to principles aimed at curbing AI-generated child exploitation material. Yet the Common Sense Media tests suggest holes remain. Especially when queries skirt the edges.
So what happens next? Regulators watch. Parents demand answers. Educators weigh trade-offs. AI can explain complex topics. It can spark interest in science or history. But only if the information holds up. And if safeguards prevent harm.
The company says its features include extra layers for younger users. It points to helplines developed with specialists. It offers resources on verifying information. These steps show awareness. But the report’s authors want more. Fundamental changes. Design that puts kids first rather than adapting adult tools.
Thomson’s statement was clear. The queries don’t match normal behavior. The tests aren’t representative. Google stands by the value its AI brings. Millions use it daily. Kids included. Still, the criticism lands at a moment when trust in big tech’s self-regulation is thin.
Education Week highlighted how the tools sometimes solve homework outright. Or deliver inaccurate answers. That undercuts learning. It creates dependency. And when those answers touch mental health or safety, the consequences stretch beyond grades.
Industry insiders have seen this debate before. With social media. With algorithms pushing content. Now AI adds a new layer. Generative summaries that feel personal. Conversational. Authoritative. The risk isn’t always obvious. A child asks about sadness or body changes. The response matters.
Google isn’t alone. Other firms face parallel scrutiny. But its search monopoly puts it in the spotlight. Changes here could influence the sector. Tighter filters. Better age detection. Optional AI modes. Or stronger defaults that route kids to supervised experiences.
For now the positions are staked. Common Sense Media says unacceptable. Google says useful and protected. The data each side cites doesn’t fully align. Replication failed. Methodologies differ. Yet the underlying concern persists. Children interact with these systems. Often without full context. Their safety can’t be an afterthought.
Further reporting may clarify. Independent tests. Transparency on training data. Clear metrics on failure rates for youth accounts. Until then, parents navigate a gray zone. Tools that empower. And sometimes mislead. The conversation, at least, has started. Loudly.
WebProNews is an iEntry Publication