THE CLAIM. ChatGPT for Teens activates automatically when a user says they are 13 to 17 or when OpenAI's systems estimate the user is under 18, wrapping teens in stronger guardrails, parental controls, and safety notifications to parents in high-risk situations.

THE CHECK. every load-bearing promise rests on two classifiers, and neither ships with a number. The age predictor has been in the works since September 2025, when OpenAI said it would 'err on the side of caution, defaulting users to the under-18 experience if it is not confident about their age', with adults offered ways to prove their age back out. The predictor rolled out in January, seven months before this launch, and no accuracy figure, false-positive rate, or evaluation has been published for it in all that live deployment. The distress detector is the same shape with higher stakes: a flagged message goes to a human reviewer, and if it qualifies, parents get notified 'in limited high-risk situations', a phrasing that quietly narrows the launch coverage's acute-distress promise. The reviewer layer is real and disclosed. The accuracy of the flagging classifier that feeds it is not, and what never reaches a reviewer never reaches a parent. Mental-health researchers have already flagged the risk of relying on automated systems in exactly these moments. The guardrails that do not need a classifier (no romantic language, break reminders, quiet hours) are sensible and real. The parts that need one are running on trust.