AI in Education AI Minute Newsroom 2026-08-18

ChatGPT now guesses whether it is talking to a child — and if it decides yes, the product changes around them

OpenAI began rolling out ChatGPT for Teens worldwide on 18 August, for 13- to 17-year-olds on both free and paid personal plans. The teen version blocks conversations about suicide and self-harm and romantic or sexual roleplay, and stops the chatbot from suggesting it has feelings, consciousness or inner experience. On schoolwork it is built to walk a student towards their own answer rather than hand one over. Teenagers are not placed there by proving their age: OpenAI does not verify ages, and instead uses what it calls age assurance to estimate from the pattern of someone's questions whether they are under 18, then routes them into the teen mode automatically. Separately, a parent and a teenager can link accounts if both agree, after which the parent can set quiet hours, block image generation, opt the child out of model training and receive alerts in high-risk situations, including signs of an eating disorder. Common Sense Media found in 2025 that more than 70 percent of American teenagers use AI chatbots for companionship.

Why it mattersFor a parent or a teacher this is the first version of the question that actually matters. Not whether teenagers use these tools — they do — but what the tool does once it has worked out who it is talking to. Two parts are worth separating. The homework design is a real change of posture, from answer machine to something closer to a tutor that withholds, and it is the part schools will feel first. The age guessing is something genuinely new: a company inferring from how you type that you are a child, and then silently rebuilding the product around that inference. It will be wrong in both directions — adults dropped into teen mode, teenagers who write like adults left outside it — and there is no counter to appeal at. Before assuming a student has guardrails on, it is worth knowing which mode they are actually in.

✓ Verified · 3 sources

ChatGPT now guesses whether it is talking to a child — and if it decides yes, the product changes around them

Related stories