OpenAI Builds a Separate ChatGPT for Ages 13 to 17

OpenAI is building age detection into the product, not just an age checkbox

OpenAI's ChatGPT for Teens is not a settings toggle. It is a distinct experience that a user is placed into, either by stating an age between 13 and 17 or by being estimated as under 18 by OpenAI's own systems. Inside it, protections are on by default rather than opt-in: reduced exposure in high-risk areas including self-harm, violence, eating disorders, and explicit material, along with restrictions on romantic language and on responses that encourage emotional dependence. It adds break reminders, sensitive-image warnings before a teen uploads a photo, a Study Mode that pushes toward working through problems rather than handing over answers, and parental controls with linked accounts and Quiet Hours.

The mechanism matters more than the feature list. A birthdate field asks a teenager to be honest. Age estimation does not ask at all. To route the right users into the safer product, the system has to form a view about how old every user is, which quietly turns age into something ChatGPT infers about you rather than something you tell it.

This is a legal product, not a feature update

OpenAI did not build an age-gated model because the market asked for one. It built it because the legal and regulatory ground shifted under the entire category. The company was named in a wrongful-death suit brought by the parents of a California teenager, reported by NBC News, and by late 2025 faced a cluster of similar filings alleging its chatbot encouraged suicide and mental-health crises. Regulators moved in parallel.

Read in that sequence, the teen tier looks less like a product roadmap and more like a compliance artifact. Every one of its headline features maps to a specific allegation or proposed rule: the self-harm limits, the emotional-dependence restrictions, the parental notifications for eating-disorder risk. This is what a company builds when the cost of an unsafe interaction has moved from reputational to litigable.

Age prediction solves one problem by creating another

To protect minors it cannot identify, OpenAI has to profile users it has no other reason to profile. That is the uncomfortable core of age estimation: child safety and user privacy pull in opposite directions, and OpenAI has resolved the tension toward inference. Guess too aggressively and adults get pushed into a restricted product they did not choose. Guess too loosely and the safeguards miss the teenagers they exist for. There is no setting that makes both groups whole.

The most revealing detail is the restriction on language that encourages "emotional dependence." That is not a content filter in the usual sense. It is an admission that the risk with teenagers is not only what the model says about dangerous topics, but the relationship the model forms with a young user over time. The companion dynamic, the very thing that makes chatbots sticky, is what OpenAI is now trying to dampen for its most vulnerable users, and the survey data shows why. In a Pew Research Center study of 1,458 US teens conducted in late 2025, 54% said they had used AI chatbots for schoolwork, but 16% used them for casual conversation and 12% for emotional support or advice. That last figure, one in eight teenagers, is the population the emotional-dependence rule is written for.

What age-segmented AI means for every other lab

When the category leader ships an age-gated tier, it stops being a differentiator and becomes a baseline. The FTC's inquiry already spans seven companies, which means age assurance is about to be a shared industry problem rather than one firm's initiative. Every consumer AI product that a teenager can reach will face the same question OpenAI just answered: how do you know who is on the other side of the conversation, and what do you owe them.

That reframes a decade of product assumptions. General-purpose AI was sold on the premise that one model could serve everyone. What OpenAI shipped this week is the first major concession that it cannot, at least not safely, and that serving everyone actually means building different products for different people behind a layer that decides which one you get.

The teen ChatGPT is not really a smaller, safer chatbot. It is the moment age stopped being metadata and became a control surface, and once the leading lab treats a user's estimated age as a reason to hand them a different model, no one building consumer AI gets to treat all users as the same user again.