Funding grants for new research into AI and teen development

OpenAI opened a grant program to fund outside research on how generative AI shapes adolescent learning, creativity, advice-seeking, and especially social-emotional development. Total pool is up to $5 million; individual awards can reach $1 million; overhead is capped at 10%. Funding comes from OpenAI Group PBC, not the OpenAI Foundation.

The brief is deliberately wide: usage patterns and outcomes; individual/social/cultural/technical moderators; safeguards and design; variance by age, culture, language, and socioeconomic context. Methods can be qualitative through experimental. Submissions run through October 6, 2026, with rolling review and notices by November 13, 2026; an interim update is expected in Q1 2027.

For a company under youth-safety scrutiny, buying an external evidence base is both public-interest work and regulatory positioning.

Signal Desk’s read: this is evidence infrastructure with a preference gradient. Review criteria openly include complementarity with OpenAI’s internal agenda and a stated preference for research involving ChatGPT, among other factors—while also weighting independence and willingness to surface findings unfavorable to AI providers. That tension is the story. Credible teen-impact science helps product and policy; preferential ChatGPT studies risk looking like sponsored literature unless methods and publication norms stay strict.

OpenAI says publication is strongly encouraged but not required for funding, and every grantee owes a final report plus recommendations where supported. That structure can accelerate useful guidance to regulators and safety teams, or it can park inconvenient results in private final reports. The independence criterion only works if selection and contracting do not quietly filter for friendly frames.

Scope choices matter. Ages 13–17 matches common product teen tiers; prioritizing countries with significant AI adoption concentrates evidence where deployment is already dense. Ethics requirements for minors—consent/assent, privacy, harm disclosure protocols, safeguarding expertise—are appropriately heavy and will slow some of the most sensitive mental-health proposals.

The likelier industry effect is a template: frontier labs funding outside adolescent research as a reputational and regulatory hedge. Watch whether Anthropic, Google, Meta, and others answer with comparable independent pots—or with tighter in-house studies that never leave the building.

Generative AI is already embedded in how many teens study, create, and seek advice. Policymakers and platforms lack a mature, shared evidence base on developmental effects, safeguard efficacy, and heterogeneous impacts across populations.