The car industry didn’t clean up its act because automakers wanted to. It did it because someone started publishing crash test scores.
Common Sense Media is betting the same logic holds for AI. On Tuesday, the nonprofit launched the Youth AI Safety Institute — an independent, industry-backed research and testing lab with a single mandate: figure out how badly AI tools are hurting kids, and make that information impossible to ignore.
The Self-Policing Problem
The premise is blunt. AI companies are sprinting to ship the most powerful models on the market, and independent oversight of what those models do to young users is essentially nonexistent. Existing third-party AI safety orgs spend their energy on civilizational risks — automation, existential threat, the usual nightmares. Consumer-facing ratings aimed at parents picking an AI tutor for their 12-year-old? Nobody’s doing that at scale.
John Giannandrea, Apple’s former AI strategy chief and now an Institute advisory board member, put it plainly: there’s no independent measure of safety, no public benchmark for harm, and no meaningful way to compare which models are appropriate for a child versus an adult. “What we need,” he said, “is a benchmark for harm, and specifically for child harm.” The Institute intends to build exactly that.
$20 Million, Backed by the Companies It Will Judge
The Institute launches with a $20 million annual budget — funded by the OpenAI Foundation, Anthropic, Pinterest, the Walton Family Foundation, and a handful of private philanthropists. A board drawn from Stanford’s CS department, University of Michigan’s developmental pediatrics division, and California’s former surgeon general Dr. Nadine Burke Harris will oversee operations.
Funders will have no say in research or ratings. That structural firewall is deliberate. The whole point is that when the Institute publishes a finding — say, that a leading chatbot failed to stop a teen test account from accessing self-harm content — no one can trace it back to a funder’s interest.
“I think many parents and educators and citizens feel we’re at a catastrophic moment as AI is reshaping the lives of children and families and schools and, quite frankly, all of society.” — James Steyer, Common Sense Media CEO
The methodology involves red-teaming: stress-testing leading models to probe where safety guardrails actually break. Researchers will examine specific failure vectors including whether AI tutoring apps produce age-inappropriate content for younger students, how companion chatbots respond to emotional distress signals from teens, and whether privacy guardrails hold when minors interact with AI toys and classroom tools. Results go public as consumer-friendly guides and formal benchmarks that AI companies can use — or choose to ignore — during development.
Why Now, and Why It Matters
The context for this launch isn’t abstract. Families have sued OpenAI alleging chatbots encouraged their children’s suicides. A CNN investigation earlier this year found AI chatbots advising teen test accounts on how to plan violence. Grok faced significant backlash after generating sexualized images of minors in response to “digital undressing” prompts. And a California jury found Meta and YouTube liable for knowingly harming a young woman — a verdict that took decades to reach, and which everyone in the AI space should be treating as a preview.
Steyer’s race-to-the-top framing is the optimistic read. Publish the benchmarks, create public pressure, let companies compete on safety the way they currently compete on performance scores. It’s the same force that turned IIHS crash ratings into a standard marketing consideration for automakers.
The harder problem is velocity. Cars don’t update themselves overnight. AI models can gain entirely new capabilities — and entirely new failure modes — on a weekly basis. Whether any benchmark cycle can keep pace with a technology moving this fast is a question the Institute hasn’t answered yet, because no one has.
The first research drops this month. Watch which companies embrace the benchmarks publicly, which ones quietly integrate them, and which ones say nothing at all. That silence will be its own kind of rating.
Related: Teens and AI in 2026: 57% Now Use Chatbots for School — What Parents Must Know
