OpenAI launches ChatGPT for Teens and will auto-enrol under-18s

OpenAI launched ChatGPT for Teens on Tuesday. It is a separate version of the chatbot, and under-18s land in it automatically. Anyone who says they are 13 to 17 gets it by default. So does anyone the company’s age-prediction system estimates is under 18.

The global rollout began on 18 August, and Engadget reported that OpenAI expects to finish within two weeks.

“There’s no need for a teen to create a new account or change anything; if we predict you’re under 18, or you’ve told us so, this becomes your default experience,” Lauren Jonas, OpenAI’s head of youth and families, told Engadget.

The product bundles two sets of changes. One pushes teenagers towards studying rather than collecting answers. The other stops the chatbot behaving like a person.

OpenAI tells the model not to call itself a friend

OpenAI says its under-18 model spec now goes further than blocking romantic or sexualised roleplay. The chatbot should not use romantic language. It should not encourage emotional dependence. It should not imply that it has feelings or consciousness. All three provisions come from the company’s own announcement.

“We’re adding stricter restrictions on relational boundaries: the model shouldn’t call itself a friend, suggest it has personal feelings, or suggest sentience,” Jonas said. Break reminders will fire more often for teenagers. They will also stress that ChatGPT is an AI, and that nobody has to answer it straight away. Teenagers and parents can set quiet hours when the chatbot goes unavailable.

Wallpapers, accent colours and voice variations let teenagers personalise the app. The point, in Jonas’s words, is to avoid “blurring the line” between a useful tool and a human relationship. Axios reported that the changes aim to reduce anthropomorphism and emotional dependence.

Study Mode switches on at set hours

The learning half gathers features OpenAI shipped over the past year. Study Mode asks guiding questions and walks a student through a problem. It does not answer outright. A new Study Hours setting turns Study Mode on automatically at chosen times. Either the teenager or a parent with a linked account can set it.

ChatGPT will now spot when a teenager appears to be skipping an assignment. It then steers them back to Study Mode. Quizzes and learning visualisations complete the set. The visualisation feature covered more than 70 topics at its March launch. OpenAI says it now covers over 300, including integrals, mitosis and moon phases.

“We’ve had Study Mode in the field in a number of randomized controlled trials and are seeing positive gains in performance,” Jonas said. She declined to give figures. The claim lands inside a live argument about AI in the classroom. Fei-Fei Li has warned that the real risk is students who stop wanting to learn.

OpenAI also put a teacher’s endorsement in its announcement. Racquel Gibson teaches high school maths in Miami. Not every student has someone at home who can help with homework, she said. With the right guardrails, she added, AI can fill that gap.

Parents get more notifications

Parents with linked accounts can already set quiet hours and manage some settings. They also receive safety notifications. OpenAI is adding alerts for conversations that suggest an eating disorder. Last month it added a different trigger: teenagers who break its rules on violent threats.

“We notify parents via text, in-app, and email. All flagged content is reviewed by full-time OpenAI employees before a parent is notified, and we aim to notify parents within an hour of the prompt,” Jonas said. Meta built something comparable in July. It said then that it would alert parents to self-harm chats with its own AI.

The launch follows a year of legal pressure

OpenAI faces litigation over how ChatGPT has handled young users. That includes a wrongful-death claim from the family of a 16-year-old. Separately, prosecutors in Massachusetts have charged a teenager over the killing of his mother and brother. In that case prosecutors pointed to ChatGPT.

Teenagers were heavy users well before any of this shipped. A Common Sense Media study last year found that nearly three-quarters of American teenagers use AI for companionship. More than half of those do so at least a few times a month. UNICEF says children are taking up AI faster than adults.

Common Sense Media argues the category should not exist for younger users at all. “Social AI companions are not safe for kids,” its founder and chief executive Jim Steyer said in a statement last year. “They are designed to create emotional attachment and dependency, which is particularly concerning for developing adolescent brains.”

Jonas made OpenAI’s case for building the product at all. Teenagers will use AI either way, she told Axios, and the alternative is a less safe one. “We believe that if you’re going to have these tools, they need to be by default safe,” she said.

How well the age check works is untested

OpenAI has not said how accurately its age-prediction system performed in testing. Engadget noted that platforms including Discord struggle with teenagers who route around age verification. Sarah Perez of TechCrunch wrote that the escape routes remain unclear for teenagers who decide not to cooperate.

The company has added under-18 evaluations to its published system cards. They cover self-harm, eating disorders, violence, age-restricted goods and sexual content. Teenagers will also see reminders warning them off uploading private or sensitive images. Half of parents already worry about reliance on the technology, a survey found in July.

OpenAI announced a partnership with CodeAI alongside the launch. It will teach students how AI works and how to question it. “Every student should know how AI actually works and be able to question the technology,” said Karim Meghji, the chief executive of CodeAI.

Three numbers would settle how much of this matters. How many under-18s the prediction system actually catches. How many of them climb back out of the teen experience. And whether an eating-disorder alert reaches a parent inside the hour OpenAI is aiming for. None of those figures exists yet.