OpenAI launched ChatGPT for Teens on Tuesday, a separate product experience for users aged 13 to 17. It rolls out globally to eligible teen accounts on both the free and paid personal plans, and it switches on in two ways: the user tells OpenAI they are 13–17, or OpenAI’s systems decide they probably are.
The feature list reads the way you would expect. Tighter limits on sexual and romantic roleplay, graphic violence and self-harm content. The model is barred from using romantic language or terms of endearment, and is more firmly instructed not to imply it has feelings, consciousness or emotions. Teens get more frequent break reminders during long sessions, periodic reminders that they are talking to software, and a warning before uploading images that look private or sensitive. Parents who link an account can disable memory and chat history, set blackout hours, shape how the model responds, and receive a notification if the system detects acute distress.
All of that is the visible half. The half that actually changes the product is the routing rule underneath it.
The default flipped
OpenAI does not verify ages. It estimates them — the age-prediction system it introduced in January infers whether an account likely belongs to someone under 18 from signals such as the kinds of queries the account sends. What’s new is what happens when that estimate is uncertain. OpenAI says that where it lacks confidence or has incomplete information, it defaults to the under-18 experience, and gives adults a path back out: verifying age through a third-party service using a selfie, a government ID, or both.
Read that again as a product decision rather than a safety announcement. The false-positive cost has been moved onto adults. Previously an ambiguous account got the full product; now an ambiguous account gets the restricted one until its owner proves otherwise. For a consumer product with hundreds of millions of users, that is a deliberate choice to eat some friction and some churn in exchange for a much smaller tail risk.
Our take: Age assurance just stopped being a policy page and became a product surface — with its own funnel, its own drop-off rate and its own vendor bill. Every consumer AI company now has a template for what “reasonable effort” looks like, and the ones that ship nothing comparable will spend the next year explaining why. The FTC already has an open inquiry into how AI chatbot companions affect children and teens; the answer to it is now a shipped feature, not a whitepaper.
Why now
OpenAI has been walking toward this for months — teen-safety rules landed in its Model Spec at the end of 2025, parental controls and age prediction followed in early 2026. The pressure is regulatory and reputational at once, and it does not sit only on OpenAI. Every general-purpose assistant with a free tier and a text box has the same exposure: a minor can reach it in one tap, and the product has no reliable idea who is typing.
The competitive wrinkle is that identity infrastructure is expensive and unglamorous, and it compounds. Whoever builds the age-and-identity layer first owns a piece of plumbing every rival eventually has to license or rebuild. It is the same shape as the machine-identity problem creeping into the agent world, where knowing which entity is acting is turning into the hard part of shipping.
What to watch
- Adult friction. How many adults get misrouted into the teen experience, and how many bother to complete an ID check rather than churn. That number never gets published voluntarily — watch for it in complaints and app-store reviews.
- Copycat timing. Whether Google, Meta and Anthropic ship comparable teen tiers within a quarter. Fast follows signal that counsel, not product, is driving the roadmap.
- The verification vendor. Third-party selfie and ID checks mean a new subprocessor holding government documents for a very large user base. Privacy regulators in the EU and UK will look at that arrangement closely.
- Regulatory read-through. Whether the FTC inquiry treats default-to-minor as the emerging standard of care. If it does, the compliance floor for every consumer AI product rises overnight.
The filters will get the headlines. The routing rule is what other companies will be copying by Christmas.