ChatGPT for Teens launches with study tools

OpenAI says its early evaluation of tools such as Study Mode has shown “promising gains in student performance”, though detailed results weren’t included.

ChatGPT can still get things wrong

Putting a chatbot closer to a student’s homework raises another familiar problem: hallucinations. OpenAI isn’t pretending ChatGPT for Teens fixes it. “We cannot eliminate mistakes, and we are clear that ChatGPT, like any AI, can be wrong,” the company told us.

Study Mode instead tries to keep students actively involved through guiding questions, explanations and knowledge checks, rather than presenting an AI-generated response as something to accept on sight. OpenAI says teenagers are encouraged to check important information against course materials, teachers and other trusted sources.

That may matter as much as generating the right answer. If generative AI becomes a routine part of education, students will need to learn that a confident-sounding answer isn’t necessarily correct.

But how does ChatGPT know you’re 15?

Every one of these protections depends on ChatGPT knowing it’s talking to a teenager, and asking doesn’t settle it. A 15-year-old can enter a different birth date or open another account.

OpenAI’s age-prediction system weighs a combination of account-level and behavioural signals:

The determination isn’t based on one conversation or a single signal, the company says, and the system is meant to identify accounts that could belong to someone under 18 even when the stated age is inaccurate or unknown.

OpenAI says it uses “privacy-preserving signals” and, where information is incomplete or uncertain, applies the safer experience by default.

The company didn’t provide a false-negative rate, which is the proportion of genuine 13-to-17-year-olds the system fails to identify and leaves in the adult experience. That number would give a much clearer picture of how far the safeguards actually reach. OpenAI also didn’t say what recourse an adult has if the system wrongly reads them as a teenager.

Stronger safeguards come switched on

Once a user is identified as under 18, additional protections apply. OpenAI says these follow the Under-18 Principles in its Model Spec, informed by developmental science and expert guidance, and cover higher-risk areas including self-harm, violence, eating disorders, dangerous activities, and explicit sexual or graphic content.

The company is also publishing under-18 evaluations in its system cards, showing how its models perform against teen-specific standards in sensitive-content scenarios spanning self-harm, eating disorders, violence, age-restricted goods and services, and sexual content.

The safeguards don’t mean ChatGPT refuses every difficult subject. That would make it far less useful for legitimate educational questions, and for situations where a teenager is genuinely looking for help. The stated goal is to reduce exposure to material that could be harmful or developmentally inappropriate while preserving the ability to learn, create and explore.

We asked whether ChatGPT for Teens had been tested against the specific conversational failure modes alleged in wrongful death suits brought against OpenAI, including the Raine case, and what changed as a result. The company restated the Model Spec principles and said that its parental controls work, without addressing the testing question. 

Parents get controls, but they can’t read the chats

For parents, one of the more interesting parts of ChatGPT for Teens is what the parental controls don’t allow.

Linked parents can manage selected settings, set Study Hours and Quiet Hours, and switch features including voice mode and image generation on or off. They cannot read their teenager’s prompts, chat history, or conversation transcripts, and they can’t monitor activity in real time. Teenagers can also delete chats without notifying their parents.

That’s a deliberate separation between oversight and privacy, and a firmer line than the phrase “parental controls” tends to suggest.

The exception is high-risk situations. A linked parent can receive a safety notification, but OpenAI says it contains only the information needed to support the teenager’s safety, not a copy of the conversation. The teenager is told when an alert has been sent.

OpenAI is expanding its safety notifications to cover eating disorders and says that asking about the subject isn’t enough to trigger one.

The notifications are reserved for what the company describes as rare, high-risk situations where activity may indicate a serious risk to the teenager. A specially trained team reviews potential alerts before sending them to confirm that the situation meets the required threshold.

When it does, the linked parent receives what’s needed to support the teenager rather than the transcript, and the teenager is notified.

There’s also a provision for teenagers with no linked parental account. In those cases, OpenAI says ChatGPT is designed to encourage support from a trusted adult or another real-world resource, without assuming that person has to be a parent.

ChatGPT doesn’t want to become your teenager’s best friend

One of the bigger changes has nothing to do with homework. OpenAI is explicitly trying to prevent teenagers from forming an unhealthy emotional dependence on ChatGPT.

Its updated under-18 Model Spec says ChatGPT should not use romantic language, encourage emotional dependence, or imply that it has feelings or consciousness.

Asked what happens when a teenager repeatedly turns to ChatGPT for support instead of a parent, teacher, counsellor or friend, the company says the system is built around holding that line.

“AI can help teens understand difficult concepts, practise new skills and build confidence, but it should never replace independent thinking, real-world relationships or support from trusted adults,” OpenAI tells HardwareZone.

When conversations lean toward isolation, exclusivity, or replacing real-world relationships, OpenAI says ChatGPT is designed to reinforce boundaries and encourage the teenager to seek offline support. Break reminders also prompt teenagers to disconnect.

When asked how emotional dependence is defined in operational terms, so that a model can act on it, and how the company measures whether the safeguard works, OpenAI didn’t say. But the words isolation, exclusivity and substitution are as specific as the answer gets.

It’s an unusual design problem. ChatGPT has to be approachable enough for a teenager to feel comfortable asking questions, while staying clearly artificial enough that it doesn’t stand in for friends, parents, teachers or counsellors. OpenAI says it continues to refine these tools using evidence and expert feedback.

But parents should consider one privacy detail. Content from teen accounts may be used to improve OpenAI’s models, depending on the account’s Data Controls. Both teenagers and linked parents can disable the model-improvement setting, and OpenAI says that when it’s off, chats and the content within them aren’t used to improve its models.

The company didn’t say whether the setting is on by default. In the parental-control imagery OpenAI supplied, the toggle appears switched on.

That imagery also referenced Sora. OpenAI says the Sora product shown there has not been available since 26 April 2026.

The bigger test starts now

OpenAI isn’t defining success purely by how many teenagers use the product. The company says it will evaluate how teenagers respond to the new features alongside Study Mode, Quiet Hours, Study Hours and its other wellbeing tools, and a planned Teen Well-Being Report will set out the framework it intends to use.

The harder test comes when teenagers start pushing the edges. Responsible Homework Reminders can steer students towards solving problems themselves, but OpenAI says they aren’t hard blocks. Age prediction can catch accounts where someone entered an inaccurate age, but the company won’t say how often it misses. Safety notifications can alert parents to serious risk while leaving conversations private, but something still has to judge when that threshold is crossed.

Singapore parents and students won’t wait long to see how that holds up. The rollout starts here today with a small percentage of accounts identified as under 18, widening over the following few weeks.