OpenAI commits $5 million to research on AI's impact on teens

This isn't coming out of nowhere. On the 18th of last month, OpenAI rolled out a dedicated version of ChatGPT for 13- to 17-year-olds, complete with a study mode and parental controls. Then on the 31st, the company announced its support for California's SB 1119, a bill aimed at AI safety for minors.

Put simply: OpenAI launched a teen-specific ChatGPT on the 18th of last month, then quickly followed up by backing California's SB 1119 youth AI safety bill. This new research grant can be seen as the next step in that sequence — an effort to gather outside evidence that could inform ongoing regulatory discussions.

OpenAI has laid out four broad research priorities. One focuses on emotional development — things like emotional regulation, resilience, self-esteem, and social bonding. Another looks at social development, covering friendships, family relationships, sense of belonging, and empathy. A third examines how outcomes vary depending on usage time, purpose, age, culture, language, and economic background. The fourth evaluates the effectiveness of safeguards, age-appropriate design, AI literacy education, and support tools for parents and teachers. OpenAI says it welcomes research teams spanning child and adolescent development, psychology, public health, human-computer interaction, communication studies, sociology, anthropology, data science, and AI safety, and that it's open to any methodology — qualitative, quantitative, experimental, observational, theoretical, or participatory.

Because the research involves minors, the requirements are strict. Applicants must specify their institutional or independent ethics review process, how they'll obtain age-appropriate consent and assent, how they'll protect participants' personal information, what protocols are in place for detecting signs of harm or crisis, whether the study involves sensitive personal, behavioral, or mental health data, and what expertise the team has in child development, clinical work, or protective practices. Handling sensitive data doesn't automatically disqualify an application, OpenAI notes, but it may require additional ethics review, specialized expertise, and more time.

Applications must be submitted in English via Google Docs and include a project overview of no more than one page, a research proposal of up to three pages, a timeline and budget, budget justification, the assumptions and risks underlying the study, an ethics and protection plan, information about the research team, and any conflict-of-interest disclosures. The research itself, however, can be conducted in any language. Reviews will weigh scientific rigor, relevance to public discourse, clarity of the proposal, likelihood of actually informing policy, products, or regulation, adequacy of ethical and protective measures, feasibility, the team's capabilities, how well the study represents diverse groups and contexts, how it complements existing evidence, and its independence — meaning the ability to reach credible conclusions even if they don't favor OpenAI.

Extra weight will go to research that uses ChatGPT itself, involves interdisciplinary teams, focuses on underrepresented populations or regions, or produces findings usable for near-term policy decisions. Eligibility is limited to applicants aged 18 or older who are affiliated with a research institution or who have relevant experience in child and adolescent development, well-being, human-computer interaction, or AI impact research; for-profit organizations are given lower priority. Selected teams can allocate up to 10% of their project budget to indirect costs, and graduate students or postdoctoral researchers are allowed to participate.

Selection results will be announced no later than November 13, and funded teams must report interim progress and early findings to OpenAI during the first quarter of 2027. Final reports must include the research questions and methods, findings and supporting evidence, limitations, and recommendations for policymakers or tech companies. OpenAI says it won't require publication in a journal or preprint as a condition of funding, but it strongly encourages teams to make their findings public. The grant program is funded and run by OpenAI Group PBC, and inquiries can be directed by email.

Researchers interested in applying can submit a project overview and proposal through the application page by October 6.

The fact that OpenAI is now funding research on top of its regulatory outreach and product changes for teen users suggests the company itself doesn't fully understand how its products affect adolescents. But since OpenAI is the one selecting and funding the studies meant to generate that evidence, how much independent weight the results carry in future policy debates will ultimately depend on what conclusions the research teams reach.