Common Sense Media Launches Youth AI Safety Institute

The nonprofit Common Sense Media launched the Youth AI Safety Institute on May 5, 2026, creating an independent research and testing organization dedicated to evaluating AI products used by children. The institute will publish open evaluations, build safety benchmarks developers can run against their models, and conduct research on youth wellbeing, according to the official press release.

Founder and CEO James P. Steyer called the effort urgent in the announcement. More than half of American teenagers now regularly chat with AI companions. Nearly a third say conversations with AI are as satisfying as—or more satisfying than—talking with real-life friends. Over half are turning to AI tools for homework help.

The institute's approach is modeled on independent crash-test ratings that show consumers whether cars are safe. The Youth AI Safety Institute will apply the same crash-test model to AI: testing the products children use most, showing parents the results, and holding industry accountable to meet a high standard of youth safety.

Technology columnist Geoffrey Fowler announced on Substack that he is joining the institute as Head of Public Engagement. He stated the initiative is backed by a $20 million annual budget (a figure that suggests serious commitment, though actual spending patterns will matter more than headline numbers).

Philanthropic funders include Lee Ainslie of Maverick Capital, Jim Coulter of TPG, John H. N. Fisher of Draper Fisher Jurvetson, Paul Tudor Jones of Tudor Investment Corp., Gene Sykes of Goldman Sachs, and the Walton Family Foundation. Industry-related funders include Anthropic, the OpenAI Foundation, and Pinterest.

The Institute is solely responsible for its standards, research, and evaluations, and maintains complete editorial independence over published results. This independence is critical given that funders include makers of technologies the institute will evaluate.

Early risk assessments already demonstrate what independent, third-party evaluation can do. The institute's website shows initial findings: Claude received a Moderate Risk rating, AI Toys received Unacceptable Risk, Grok AI received Unacceptable Risk, Gemini K-12 received High Risk, and AI Chatbots for Mental Health Support received Unacceptable Risk.

These assessments focus on AI's approach to interaction in two key areas: Technical (Safeguards, Privacy, Content Safety, Parental Controls) and Developmental (Appropriateness, Age-Appropriate Design, Learning Impacts, Social-Emotional Effects). The physical reality of this work means clicking through chatbot interfaces, testing age-gating mechanisms, and measuring how long it takes a child to accidentally access inappropriate content.

The Institute will be guided by a Board of Advisors composed of distinguished experts in AI, youth development, child safety, mental health, and education. Advisors include Dr. Nadine Burke Harris, pediatrician and former Surgeon General of California; John Giannandrea, former SVP of Machine Learning and AI Strategy at Apple and Chief of Search & AI at Google; John King Jr., Chancellor of SUNY and former U.S. Secretary of Education; Dr. Jenny Radesky, Associate Professor of Pediatrics at University of Michigan Medical School; Mehran Sahami, Professor at Stanford University; and Dr. Vivek Murthy, former Surgeon General of the United States.

CNN reported that John Giannandrea joined the institute's advisory board and argued for independent measures of model appropriateness for children. Euronews reported that former European Commission executive vice-president Margrethe Vestager publicly backed the initiative at a Copenhagen event.

The Institute is working alongside a growing network of strategy, research, and technical evaluators, including established partnerships with Transluce, Humane Intelligence, and Stanford Medicine's Brainstorm Lab for Mental Health Innovation.

For practitioners, this push could create publicly visible benchmarks for model behaviour in youth-facing contexts. New evaluation artifacts—open-source tests, safety criteria—may become requirements product and safety teams need to consider when designing or certifying family- and child-oriented features. Comparable third-party testing programs in other sectors have pressured vendors to change defaults, improve transparency, or provide differentiated products for sensitive populations.

Industry reporting notes practical difficulties: AI systems update frequently, behave differently across contexts, and resist standardised test conditions, which complicates direct analogies to automotive crash testing. The institute has not yet explained what a "crash test" looks like for chatbots and similar products in detail.

Observers should track three implementation choices that will determine the institute's practical influence: the specificity of age-grade safety criteria, how tests handle continuously updated models and personalised behaviour, and whether results are actionable for regulators or commercial partners. John Giannandrea noted the lack of independent measures for determining which models are age-appropriate, which underscores demand for standardized tests.

Also watch for published methodologies and open-source test suites from the institute and for whether major platform vendors participate, decline, or adapt based on published rankings. The institute's outputs—if it makes methodologies public and gains broad industry engagement—could become de facto requirements in product risk assessments for youth-facing features.

The funding structure creates an inherent tension: industry funders include companies whose products will be graded. Whether the institute can maintain credibility while accepting money from the very entities it evaluates remains the real question.

Absent clear, reproducible test protocols and buy-in from platform vendors, the initiative's influence will depend on the visibility and perceived credibility of its early reports. The crash-test analogy works well in marketing materials, but AI doesn't break down in predictable ways like metal and glass.

The Youth AI Safety Institute represents a notable experiment in applying consumer-facing accountability methods to AI used by children. Whether it changes behavior or just generates press releases depends on execution details that haven't been fully disclosed yet.

Time will tell if this works. For now, parents can bookmark the institute's website and hope the first round of evaluations actually changes something before their kids spend another weekend talking to chatbots that don't know they're talking to children.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn