The first AI lab to put real money behind measuring how their own tech affects kids under 13.

The Summary

The Signal

OpenAI isn't just signing another safety pledge. They're funding third-party research to build measurement frameworks that don't exist yet. The $2 million commitment targets something the AI industry has mostly avoided: empirical data on how conversational AI affects cognitive development, social learning, and emotional regulation in kids under 13.

The void here is real. We have decades of research on screen time and social media. We have almost nothing on what happens when a 10-year-old uses an AI that can tutor them in math, help with creative writing, and respond to emotional prompts. Parents are guessing. Schools are making policy in the dark. And the labs building these systems have zero baseline metrics.

"The first AI lab to put real money behind measuring how their own tech affects kids under 13."

What makes this different from typical tech safety theater: the focus on shared measurement tools. Not internal safety teams. Not age verification. Not content filters. Shared tools means portable frameworks that work across labs, researchers can replicate, and regulators might actually trust. OpenAI is betting that standardized child safety metrics serve everyone better than each lab building proprietary ones.

The research priorities signal where OpenAI sees gaps:

  • Developmental impact across age ranges (a 7-year-old and a 12-year-old are different species)
  • Comparative risk/benefit analysis (is AI tutoring better or worse than YouTube for a struggling reader?)
  • Longitudinal data (does daily ChatGPT use at age 11 change anything measurable at age 14?)

The Implication

If this produces actual measurement tools, expect them to become the basis for AI child safety regulation worldwide. Governments don't write laws without metrics. Metrics don't exist without funding. OpenAI just bought a seat at the table where those metrics get defined.

For parents and educators: this research won't answer "should my kid use AI" for at least two years. But it might answer "how much, doing what, with what guardrails" by 2027. Watch for the research grantees and early findings.

Sources

OpenAI Blog