Routing Runs on Prediction, and Uncertainty Resolves Strict
There are two ways into the mode: state that you are 13 to 17, or be estimated as under 18 by OpenAI's age-prediction system. The prediction draws on behavioral signals including chat content and time of day. What matters is the default direction — when information is incomplete or confidence is low, the user lands in the safer experience rather than being waved through. Adults caught by a false positive can verify with a government ID and a selfie, handled by the third-party vendor Persona, which deletes the upload within seven days; OpenAI never receives the ID or the selfie.
More Than a Longer Blocklist
The content tightening covers self-harm, eating disorders, violence, and sexual or romantic roleplay. The more interesting changes sit elsewhere. The model is barred from using terms of endearment with teens and more strongly instructed not to imply it has feelings or consciousness. Voice mode can be switched off. A break reminder fires roughly every 90 minutes, noting that the user is talking to an AI. None of that targets "said the wrong thing" — it targets emotional dependence itself. Schoolwork gets additions instead of restrictions: Study Mode supplies guiding questions and step-by-step support, and homework reminders appear when a teen looks like they are shortcutting an assignment rather than handing over the answer. On the parent side there are Study Hours and Quiet Hours, plus safety notifications in a limited set of high-risk situations — reviewed by trained personnel before being sent, per OpenAI. Parents cannot read the conversations, a limit several outlets flagged as the catch in these controls.
Why Now
OpenAI says nearly nine in ten teen users come to ChatGPT for learning, information, skill-building or productivity (OpenAI's own figure, not independently verified). The other half of the context is litigation: several suits over protections for young users are ongoing, including one brought by the family of a 16-year-old who died by suicide. For teams building education products or anything aimed at younger users, the practical implication is at the interface: if your product uses ChatGPT as a backend, age determination for your end users may now be made on OpenAI's side, and the behavioral constraints move with it. Test the output difference under a teen account before you ship.
via: TechCrunch, Axios, OpenAI Help Center: ChatGPT for Teens, OpenAI: Our approach to age prediction