OpenAI's $5m teen AI research fund, and what it quietly admits

Every parent with a teenager and a chatbot in the house has asked some version of the same question. Is this good for them or not? On 8 September, OpenAI said, in effect, that it would like to know too.

The company is putting US$5 million into independent research on how generative AI affects the lives and development of young people aged 13 to 17. Individual grants go up to US$1 million, applications close on 6 October, and successful teams hear back on or before 13 November. It is a modest sum for a company of that size, but the announcement is worth reading closely, because it says plainly what the industry does not yet know.

The admission in the first paragraph

OpenAI’s own framing is careful. Generative AI, it writes, “is becoming part of how teens learn, create, and seek advice. Its effects are likely to be nuanced.” The stated aim is “to build a stronger evidence base about the way teens use AI, as well as the impacts of AI on their lives.”

Read that next to the calendar. ChatGPT for Teens, the separate under-18 version with Study Mode, homework reminders and stronger content protections, launched on 18 August. (We covered what changed in ChatGPT for Teens at the time.) OpenAI does cite research for that product: learning science on active retrieval and self-explanation, and what it calls “promising gains in student performance” from its own early evaluation of tools like Study Mode. That is a reasonable basis for a study feature. It is not the same as knowing what months of daily chatbot use does to a 14-year-old’s friendships or moods.

That is the gap the grants are aimed at. The research priorities read like a list of what parents actually lie awake over: emotion regulation, resilience, self-esteem, friendships, family relationships, and “the relationship between AI and human connection.” Another strand asks whether safeguards, AI literacy and parental support actually work.

The product came first. The evidence is being commissioned now. That is not a scandal, it is how most technology reaches children, but it is worth saying out loud.

[[photo:1]]

What the fine print gets right, and what it leaves open

Some of it is genuinely good. Anyone studying minors has to explain their ethics review, how they will handle informed consent and age-appropriate assent, how data will be protected, and, importantly, “how researchers will respond to disclosures of harm or imminent risk.” Applicants must declare industry funding and other conflicts. One of the review criteria is a project’s ability “to produce trustworthy findings regardless of whether those findings are favorable to providers of AI products.”

Three details deserve a raised eyebrow, though. This is our reading, not a claim that anything improper is planned.

First, the output. Every funded project must give OpenAI a final report. OpenAI “strongly” encourages public release, then adds: “publication will not be a condition of receiving funding.” So a study can finish as a private document on the funder’s desk.

Second, the reviewers. Applications go to “a panel of internal experts and advisors,” reviewed on a rolling basis.

Third, the preferences. The call says preference may go to research involving ChatGPT, and one criterion is how well the work “complements OpenAI’s internal research agenda.” Studying the tool children here use most makes sense. It also means the company helps choose which questions about its own product get asked.

None of this makes the research worthless. It does mean the findings to trust most will be the ones that end up public, with methods attached.

Singapore’s children start earlier than the fund does

Here is where the local picture sharpens things. The fund covers ages 13 to 17. Singapore’s first national look at children’s AI use suggests the habit starts well before that.

The study, by Dr Xuejiao Chen of A*STAR and Professor Jean Yeung of NUS, draws on nationally representative data from 2,985 children aged eight to 13 in the Singapore Longitudinal Early Development Study (SG-Leads), and was presented at NUS on 21 May. As The Straits Times reported, more than half of eight-year-olds had used AI tools, more than seven in 10 had by age 10, and more than nine in 10 by age 13. ChatGPT was the most popular, with Meta AI and Google Gemini also widely used.

The researchers’ next question is almost the same as OpenAI’s. Prof Yeung said the next step would be to investigate how these AI behaviours relate to children’s achievement outcomes or social emotional outcomes. Dr Chen was blunter about where families stand today: “I think a lot of parents do not understand how AI works, or how to safeguard their children when they use AI.”

So by the time a Singapore child enters the age range OpenAI is paying to study, they may already have five years of chatbot habits behind them. Those earlier years are outside this fund’s scope.

[[photo:2]]

What to do while the studies run

Interim updates are due in the first quarter of 2027. Solid answers will take longer, and your child is using these tools this week. A few things follow.

• Treat “promising” as early. When a company, including ours, says a study feature helps learning, ask who measured it and whether the numbers are public.
• Watch for the published versions. When grant results appear next year, look for a paper or public report with methods, not a summary in a company blog post.
• Be the local study. The topics on OpenAI’s list, friendships, family relationships, well-being, are things you can notice at home long before any journal reports on them. Is your child still seeing friends? Still bringing problems to you?
• Start the talk before 13. Singapore’s own data says the tools arrive in lower primary. Dr Chen’s point about AI literacy applies to parents as much as to children.

A lab paying to find out what its product does to teenagers is a good thing. It is also a reminder that the honest answer to “is this good for them?” is, for now, “nobody fully knows yet”. Parents can act on that before the reports come in.

Disclosure: Mentus AI makes an AI mentoring app for children, so we have a commercial interest in how families judge AI learning tools. We are not involved in OpenAI’s grant programme.