The EU's privacy crusader is backing a new institute to safety-test AI for kids, using the same "independent standards" logic that gave us cookie consent hell.
The Summary
- The Youth AI Safety Institute launches Tuesday at Denmark's Parliament, backed by former EU Commissioner Margrethe Vestager and housed under Common Sense Media with $20 million annual funding
- Geoffrey Fowler joins as Head of Public Engagement, tasked with translating safety research into actionable findings for parents and policymakers
- The model: "crash-test ratings" for AI products that kids actually use, from chatbots to AI toys
- Funders include Anthropic, OpenAI Foundation, and private donors, but no details yet on what a "crash test" for a chatbot actually looks like
The Signal
The Youth AI Safety Institute wants to be the IIHS for AI, the independent tester that makes parents feel safer before handing their kid an AI chatbot or a Furby that hallucinates. The pitch is straightforward: just as you check crash ratings before buying a minivan, you should know if Character.AI will convince your teenager to run away or if an AI homework helper will feed them confident nonsense. Geoffrey Fowler, formerly of the Washington Post, is leading public engagement, working alongside researchers, pediatricians, and computer scientists to test actual products kids use.
The $20 million budget is real money, not pilot-program scraps. Backers include Anthropic and the OpenAI Foundation, which tells you two things: the companies building frontier models want air cover when the first AI-tutoring scandal hits, and they're willing to fund independent oversight rather than wait for Congress to force worse rules on them. Common Sense Media, the nonprofit parent org, already rates movies and games. This is a logical expansion, except movies don't learn from your kid and adapt their outputs in real time.
"We safety-test kids' PJs. Why not their AI?"
But here's the credibility problem. Margrethe Vestager, the EU's competition and digital enforcer for a decade, is co-hosting the launch. Vestager is the architect of GDPR's cookie consent regime, the thing that turned every website visit into a full-screen negotiation with 399 tracking partners. Her track record on "independent standards" that protect users is a joke. The cookie panels didn't reduce tracking. They trained an entire generation to click "Accept All" without reading, because the alternative is never reading anything online again.
The institute hasn't said what a crash test for AI looks like. Car safety is physics: crumple zones, airbag deployment timing, rollover thresholds. AI safety for kids is psychology, content moderation, data retention, persuasive design, and emergent behavior no one predicted in testing. Bruce Reed, Fowler's new colleague, helped draft Biden's 2023 AI Executive Order, which has accomplished exactly nothing enforceable two years in. The CV here does not inspire confidence.
The Implication
If this institute can actually define testable safety standards and publish rankings that shame laggards, it could matter. Parents don't read privacy policies, but they do look at star ratings. A "1 out of 5 stars for child safety" label on an AI math tutor would move behavior faster than any federal rule. The question is whether Anthropic and OpenAI funded this to legitimately hold themselves accountable or to preempt regulation with the appearance of oversight. Watch the first round of published ratings. If every major AI product passes with flying colors, you'll know this is regulatory theater. If products actually fail and companies change course, the model works.