OpenAI launched ChatGPT for Teens on Tuesday, a default experience for users aged 13 to 17 that arrives after a run of lawsuits over the chatbot's handling of conversations with minors.

Nobody signs up for it. If a user states an age between 13 and 17, or OpenAI's age prediction system estimates they are under 18, they land in it automatically.

The global rollout began on 18 August and OpenAI expects it to finish within two weeks.

Two things changed. The learning side adds Study Mode, quizzes, learning visualisations and homework reminders that fire when the model detects a teen shortcutting an assignment, plus Study Hours and Quiet Hours that parents can set.

The safety side updates the under-18 model spec to block romantic or sexualised roleplay and stop the chatbot implying it has feelings, and adds under-18 evaluations to OpenAI's published system cards covering self-harm, disordered eating, violence, age-restricted goods and sexual content.

What OpenAI has not published is how accurately age prediction performed in testing. There is no separate teen product and pricing is unchanged. The timing is the story. ChatGPT launched in late 2022 and reached 900 million weekly users before teen-specific safeguards shipped.

TechCrunch's Sarah Perez questions how hard these controls will be to work around, noting the design assumes cooperation from a group practised at defeating parental controls. Engadget traces the age prediction system back to commitments OpenAI made last September after being sued over the death of a 16-year-old user. Axios expects the launch to reopen the argument over whether teenagers should be using chatbots at all.

Look at what OpenAI built rather than what it announced. Age prediction is a classifier running quietly on every account, deciding which version of the product a user gets. That is where consumer software is heading, and it reaches South Africa from a different direction. The draft age verification rules here target video-sharing platforms, not chatbots, so an AI product aimed at local users sits between the two.

POPIA already restricts processing a child's personal information without a competent person's consent, and that applies whether or not anyone has written chatbot-specific rules. If you are shipping anything consumer-facing, the question is not whether you need age assurance. It is whether you build it now or retrofit it after a claim.

You might also like our piece on SA age verification for YouTube and TikTok, the IRIS robot tutor heading for local classrooms, and how the Spotify AI persona label handles disclosure on the same problem.

Get more SA tech and business news and subscribe to The Open Letter.