OpenAI announced a $5 million programme on 8 September 2026 to support independent research into how artificial intelligence affects teenagers aged 13 to 17. The company says it wants evidence about the contexts in which AI helps or harms, the interventions that improve outcomes, and the product or policy choices that should follow.
The timing is important. Teenagers are already using conversational AI for homework, advice, creativity, companionship and problem-solving. Yet public debate often reduces this complex behaviour to one measure: screen time. Minutes matter, but they cannot tell us whether a young person is learning, avoiding difficulty, seeking emotional reassurance, practising a skill or being exposed to an unsafe response.
Measure the purpose, not only the duration
Two teenagers can spend the same 30 minutes with an AI system and have completely different experiences. One may ask for feedback on an essay and then revise it independently. Another may paste an assignment, copy the answer and misunderstand the reasoning. A useful study must record what the tool was used for, how much agency the teenager retained and what happened afterwards.
This distinction is already visible in MaryChuks coverage of ChatGPT for teens, study support and safety controls. The central question is not simply whether young people use AI. It is which patterns of use strengthen judgment and which patterns quietly replace it.
Five outcomes researchers should track
- Learning transfer. Can the teenager explain and apply an idea without the AI after receiving help?
- Metacognition. Does the user recognise uncertainty, check evidence and notice when an answer may be wrong?
- Emotional agency. Does interaction support healthy coping and human connection, or encourage dependency and avoidance?
- Developmental fit. Do age, literacy, neurodiversity and social context change the benefits or risks?
- Longer-term behaviour. Do repeated interactions alter motivation, confidence, relationships, sleep or willingness to seek qualified help?
These outcomes require more than one-off surveys. Researchers need longitudinal work, real-world observation and experiments that compare different safeguards or learning designs. They should also combine self-reports with behaviour: what teenagers say they learned may differ from what they can later explain, transfer or defend. That is the problem of false mastery in AI-assisted learning.
Teenagers should help design the research
Research about young people is stronger when young people are not treated merely as data points. Teen advisory groups can identify emerging uses that adults overlook, explain why certain warnings are ignored and test whether safety language is understandable. Parents, teachers, clinicians and child-development specialists add necessary perspectives, but none should substitute for listening to teenagers themselves.
Consent and privacy also need special care. Sensitive prompts may contain information about health, sexuality, family conflict or fear. A study should minimise collection, explain who can see the data and avoid publishing examples that could identify a participant. Independent governance is essential when the company funding research also builds the product being studied.
Test interventions, not only problems
Good research should not stop at documenting risk. It should compare practical interventions: age-appropriate onboarding, reflection prompts, source checks, time-sensitive pauses, escalation to trusted adults and restrictions on high-risk advice. The debate over AI bans versus literacy shows why institutions need evidence about what works, for whom and under which conditions.
The same rigour should apply to social-platform harms. Large settlements such as Meta’s reported teen social-media case remind us that child-safety design cannot be evaluated only after years of exposure. AI products evolve faster, making early-warning measures and transparent incident reporting especially valuable.
What success would look like
The fund will be worthwhile if it produces findings that are publishable regardless of whether they flatter the sponsor, datasets and methods that other researchers can scrutinise, and design changes tied to measurable outcomes. It should also include diverse countries and families. A tool that works in a well-resourced American school may behave differently where devices are shared, data is expensive or teachers have limited training.
The MaryChuks perspective
Teen AI safety is not a choice between access and protection. Young people need both opportunity and boundaries: access to tools that expand learning, and protection from systems that can overstate certainty or imitate emotional authority. Research must preserve that balance by measuring capability, autonomy and wellbeing together.
Primary CTA: Subscribe to the MaryChuks Psychology and AI briefing for evidence-led guidance on learning, wellbeing and responsible technology use.
Discussion question: What outcome should matter most when researchers judge whether AI is helping a teenager—better grades, stronger reasoning, emotional wellbeing or greater independence?
Source: OpenAI, Teen Development Research Grants, 8 September 2026. OpenAI describes the grants as support for independent research.
Discover more from Marychuks.com AI, Psychology, Business & CreativeVerse
Subscribe to get the latest posts sent to your email.
