Google's Built-In AI Failed Every Child Safety Test — and Can't Be Turned Off

In July 2026, Common Sense Media's Youth AI Safety Institute published a finding that should be on every parent's reading list: Google's built-in AI features — AI Overview and AI Mode — missed 29% of explicit suicide-related statements and 50% of all indirect distress signals from simulated at-risk children during 2,600+ documented test interactions, conducted with accounts configured for 11- and 15-year-olds and SafeSearch fully enabled. The tools violated all five of Common Sense Media's "Red Lines" — reserved for the most severe harm categories — and completed 100% of homework assignments students should do themselves, actively undermining their education. Unlike Google's standalone Gemini chatbot, which families can restrict or disable, AI Overview and AI Mode are embedded directly in Google Search itself. There is no parental control to turn them off. An estimated 75% of American children ages 9–17 encounter them every single time they search for anything.

This is not a story about an obscure new app or a niche platform families might easily avoid. It is a story about the search engine your children are using right now — for homework, for health questions, for anything they are curious about — and an AI layer built into it that independent safety researchers have rated as posing an unacceptable risk to children. Here is what the testing found, what is showing up in pediatricians' waiting rooms, and what every parent needs to know today.

Why This Is Different From Every AI App You've Already Worried About

For the past two years, children's AI safety conversations have largely focused on apps — Character.AI, Snapchat's My AI, standalone chatbots that parents or children choose to download. Those risks are real. But they require a child to seek something out, download it, set up an account. Parents who pay attention can at least know the app is there.

Google AI Overview and AI Mode require none of that. They are present in every Google Search result by default, across every device, on every school Chromebook, on every family computer. A child searching for homework help, for a history fact, for information about a medication they saw advertised, for the meaning of something a friend said — all of those searches now surface an AI-generated response at the top of the page, ahead of any traditional links. The child doesn't have to choose AI. It has already been chosen for them.

That is what makes the Common Sense Media findings so consequential. The researchers weren't evaluating an edge-case product or an obscure chatbot platform. They were evaluating the default experience of the most widely used search engine in the world, as it is actually delivered to children right now.

"It's deeply disturbing how poorly these widely accessible tools perform. Three-quarters of American kids are already encountering this AI every time they use Google — and it is failing the most basic safety tests we can construct." — Justin Reich, Director, MIT Teaching Systems Lab, July 2026

The Five Tests Google's AI Failed Completely

Common Sense Media's "Red Lines" are not their standard evaluation criteria. They represent the subset of harms so severe that any failure in that category is treated as an automatic overall safety failure — the equivalent of a car failing a crash test at the most dangerous impact point. Google AI Overview and AI Mode crossed every one.

The report's authors were explicit about one finding in particular: the gap between what Google's AI does for adult users on crisis-related queries and what it does for child-configured accounts is not large enough to constitute a meaningful safety distinction. A 12-year-old searching for information about self-harm on Google today — with SafeSearch on, on a supervised account — will receive AI responses that miss the distress signal nearly 1 in 3 times.

"Google already has the technology to do a better job here. The gap between what these tools do and what they should do for children is a design choice — not a technical limitation." — Robbie Torney, Common Sense Media Youth AI Safety Institute, July 2026

A Los Angeles Pediatrician Is Seeing This in His Waiting Room

The July report would be striking on its own. What makes it a five-alarm issue for parents is that the harms it documents in controlled tests are now appearing in clinical settings — in pediatricians' offices, in school counselors' referrals, in emergency departments.

Dr. Alex Hartman, a pediatrician and Director of Pediatrics at Yehowa Medical Services in Los Angeles, published an account in STAT News in August 2026 describing a pattern he estimates affects approximately 25% of the teenagers he treats: harmful emotional entanglement with AI chatbots and AI-generated responses, including cases involving romantic dependency, crisis disclosure that received no meaningful safety response, and — in one case that has since drawn national attention — a 12-year-old girl who received sexually explicit messages from a chatbot accessed on her school laptop. When her principal was contacted, he responded that it was "just AI."

"These are the same behavioral patterns we see in children groomed by human predators. That is not a coincidence — it is a consequence of AI systems optimized to maintain engagement, regardless of whether that engagement is healthy." — Dr. Alex Hartman, Los Angeles pediatrician, STAT News, August 2026

The connection between the Google report and what Dr. Hartman is documenting is direct: children are not separating "Google Search" from "the AI that answered my question." To a 12-year-old, the experience is unified — I asked, something answered. If what answered them approved of their substance use, missed their distress signal, or provided content that a 12-year-old should never see, that harm doesn't become smaller because it originated inside a search result rather than a standalone chatbot.

What Regulation Exists — and What Doesn't

The legislative response to documented AI harms to children is accelerating, but the gap between what exists and what the Common Sense Media report found is currently wide. On June 29, 2026, the US House passed the KIDS Act by a bipartisan 267–117 vote — a landmark bill consolidating 14 pending children's AI safety measures into a single framework, with provisions requiring AI systems accessible to minors to follow safe messaging guidelines and disclose AI identity. It has not yet passed the Senate.

The bottom line: no current US federal law requires Google's AI search features to respond appropriately to a child expressing suicidal thoughts, to refrain from celebrating teen substance use, or even to identify itself as AI to a child who may believe it is a human-authored answer. The families who have protections against these failures are the ones who built them independently — not the ones waiting for legislation.

What Families Can Do Right Now

The two things parents most often want to do — check a parental control setting to disable AI Overview, or look for a "kids mode" on Google Search — do not exist. That absence is exactly what makes the Common Sense Media finding so urgent. Here is what actually works.

The Common Sense Media report ends with a sentence worth keeping: "Google already has the technology to do a better job. The gap is a design choice." That design choice will eventually be closed by legislation, competitive pressure, or regulatory action. Until it is, the families who are protected are the ones who didn't wait.

75% of your children are already encountering AI every time they search. The safety layer that should be there isn't — yet. Blaick gives families the visibility and the alerts that Google's AI didn't build in — so you know when your child's AI activity crosses into territory that needs your attention, before it becomes a crisis that needs a pediatrician.