OpenAI Launches ChatGPT for Teens: Age Prediction, Quiet Hours, and a Chatbot That Refuses to Do Your Homework

On Tuesday, August 18, OpenAI launched ChatGPT for Teens, a dedicated version of its consumer chatbot that switches on automatically for anyone who registers as 13 to 17 years old — and, more significantly, for anyone whose account its age-prediction system estimates belongs to someone under 18. The launch is OpenAI’s most aggressive attempt yet to reconcile two competing pressures: teenagers are among ChatGPT’s heaviest and fastest-growing user groups, and they are also the group regulators, parents, and researchers worry about most.

The result is a product that is part safety wrapper and part pedagogy engine. It blocks entire categories of conversation, hands parents a set of controls that stop deliberately short of surveillance, and — in its most interesting design decision — tries to make the chatbot slightly worse at doing your homework on purpose.

How teens get routed in

Nobody has to verify their age to use ChatGPT for Teens. The experience applies itself automatically when a user states an age between 13 and 17, or when OpenAI’s age-prediction system — rolled out at the start of 2026 — estimates an account belongs to a minor. That system looks at account signals such as the general topics discussed, usage times, and how and when the account is used. OpenAI acknowledges it makes mistakes: an adult wrongly sorted into the teen experience can get out through Persona, a third-party verification service that may request a live selfie, government ID, or both depending on the country. OpenAI says it only receives the age confirmation, and Persona deletes the selfie or ID within seven days.

The framing matters legally as much as technically. This is an age-assurance system rather than a mandatory identity check, a distinction designed to thread the needle between children’s-safety regulation (which keeps tightening) and privacy law (which frowns on collecting IDs from children in the first place). Users under 13 remain banned outright from the platform.

What gets blocked, and what parents actually see

Inside the teen experience, the guardrails tighten across the board. Romantic and sexual roleplay is blocked outright, as are romantic-companion dynamics more generally — OpenAI says the product will avoid romantic language, discourage emotional dependence, and avoid implying it has feelings or consciousness. Depictions of self-harm, graphic violence, and dangerous activities are restricted, and the system surfaces warnings and safety information around topics like eating disorders rather than engaging with them as ordinary conversation topics.

The parental side is deliberately narrow. Parents do not get access to their teen’s conversations. After the teen accepts an account-linking invitation, a parent or guardian can set quiet hours that lock access to ChatGPT entirely, disable voice mode, turn off memory, remove image generation, and opt the teen’s data out of model training. They can also receive safety notifications — but only in “limited high-risk situations”: a potential self-harm flag goes to a specially trained review team, and if confirmed, parents may be contacted by email, text, and push notification unless they’ve opted out. OpenAI is expanding notifications to cover some eating-disorder situations and alerts when a linked account is banned for violent activity, while explicitly excluding ordinary fiction writing, gaming, news, politics, and general anger from triggering notices.

That narrowness is a feature, not an oversight — but it has known limits. A January 30, 2026 academic preprint that tested OpenAI’s parental-control system found the notifications were selective rather than comprehensive: several tested categories, including fraud, hate speech, and malware, produced no alerts at all, while physical-harm, pornography, and some health prompts triggered alerts only intermittently. The study examined an earlier configuration and doesn’t measure this week’s launch, but it explains why OpenAI is careful to describe the alerts as safety interventions rather than a monitoring system, and why Common Sense Media’s Robbie Torney called OpenAI’s parental controls “a good starting point” that works best combined with family rules and active parental involvement.

Homework help with built-in friction

The pedagogical half of the launch is where OpenAI is breaking the most new ground. Study Mode — guiding questions and step-by-step explanations rather than direct answers — now comes with supporting features: quizzes, “learning visualizations,” and Study Hours that keep Study Mode locked on during designated periods.

The headline addition is the Responsible Homework Reminder. OpenAI says the system is designed to recognize when a teen appears to be trying to shortcut an assignment — fishing for a finished answer rather than help understanding — and to redirect the interaction toward collaborative problem-solving instead of simply producing the solution. In effect, OpenAI has built a chatbot that is trained to notice when it’s being used as a cheating engine and to decline to play along gracefully. A high school math teacher quoted in the announcement, Racquel Gibson of Miami, framed it as a way for students to understand their mistakes and keep working when no adult is around to help.

OpenAI also announced a partnership with CodeAI focused on helping students and educators understand how AI works, question its outputs, and use it in classrooms with teachers involved — a companion effort aimed at the institutional side rather than the living-room side.

Consolidation, not revolution — and that’s the story

The Verge’s Robert Hart makes the sharpest observational point about the launch: most of these protections aren’t new. Age prediction arrived in January, parental controls and study mode roughly a year ago, and more frequent break reminders for teens last month. What’s new is the packaging — a named, coherent “ChatGPT for Teens” brand that consolidates scattered safeguards, adds a handful of features (quiet hours, homework reminders, expanded alert categories), and signals a durable product commitment rather than a settings menu.

That consolidation is itself the news. For two years, the industry’s teen-AI story has been reactive: platform launches first, safety features bolted on after criticism. OpenAI is now treating minors as a first-class product surface with its own announcement, its own help documentation, and — notably — new under-18 evaluations published in its system cards covering self-harm, eating disorders, violence, age-restricted goods, and sexual content. What OpenAI has not disclosed is a launch-wide accuracy rate for the age-prediction system or the new safeguards, which remains the most consequential open question: an age-assurance system that misses minors is a privacy win for adults and a protection failure for exactly the people this product is meant to shield.

With AI chatbots now a default study companion for a generation of students, the stakes of getting this right extend well beyond OpenAI. ChatGPT for Teens is best understood as a live experiment in whether “safe by default, private by design, and deliberately unhelpful at cheating” can be a viable mass-market product — and whether the rest of the industry copies the template.