OpenAI is rolling out ChatGPT for Teens, a version of its assistant that automatically kicks in when the system estimates a user is under 18 or someone states they’re between 13 and 17. The pitch is straightforward: teens get AI help without the usual risks, and parents get controls that didn’t exist before. It builds on a stack of prior work — parental controls, an under-18 model spec, age prediction — but this launch adds something genuinely new: homework reminders that flag when a student seems to be shortcutting an assignment and redirect them toward step-by-step problem solving instead of handing over answers.
The learning toolkit is the real story here. Study Mode, which OpenAI says was built with teachers, scientists, and pedagogy experts, uses guiding questions, scaffolding, and metacognitive prompts to push students toward comprehension rather than copying. There are quizzes, learning visualizations, and Study Hours — a feature that lets teens or parents set times when Study Mode turns on by default. OpenAI’s early evaluations of Study Mode show “promising gains in student performance,” though they’re not sharing the numbers yet. The design leans on learning science showing active retrieval and self-explanation beat passive review for retention.
OpenAI also announced a partnership with CodeAI aimed at teaching students how AI works, how to question it, and how to create with it — with teachers shaping how the tech enters classrooms. The company was careful to position ChatGPT for Teens as an out-of-classroom tool, keeping ChatGPT for Teachers as the separate, institution-managed product for districts. That split matters: it sidesteps the messier questions about AI inside schools while still claiming the education space.
Teens are already building real things with these tools. In Virginia, three students used ChatGPT to develop WiFind, a search-and-rescue system that uses Wi-Fi signals to locate survivors under rubble. In Texas, Crystal Yang built Audemy, an audio game platform for blind students, scaling it to 200,000 users and 50 games. Those aren’t homework hacks — they’re the use cases OpenAI wants to normalize. Whether anti-cheating detection actually works at scale is the open question. Detecting intent to shortcut is hard, and false positives could frustrate exactly the students this is meant to help.
