OpenAI shipped a version of ChatGPT on Tuesday that decides for itself whether the person typing is a child.
ChatGPT for Teens began rolling out globally on August 18, and the company expects the rollout to finish within two weeks. It is a distinct product experience for users aged 13 to 17, bundling tighter content defaults, learning tools built to stop students outsourcing their homework, and optional parental controls. What makes it unusual is not the feature list but the enrollment mechanism: anyone who tells OpenAI they are 13 to 17 lands in it by default, and so does anyone the company’s age-prediction model estimates is under 18.
“There’s no need for a teen to create a new account or change anything; if we predict you’re under 18, or you’ve told us so, this becomes your default experience,” said Lauren Jonas, OpenAI’s head of youth and families.
The safety half of the product tightens defaults around self-harm, suicide, eating disorders, graphic violence, dangerous viral challenges, and sexual or romantic role play. The more novel restrictions govern how the model talks about itself. Under OpenAI’s updated Under-18 Principles, ChatGPT is barred from using romantic language or terms of endearment with teens, instructed not to encourage emotional dependence, and told more forcefully not to imply that it has feelings or consciousness.
“We’re adding stricter restrictions on relational boundaries: the model shouldn’t call itself a friend, suggest it has personal feelings, or suggest sentience,” Jonas said. Break reminders fire more often for teenagers and explicitly restate that ChatGPT is an AI and that nobody has to reply right away. Teens also get warnings before uploading potentially sensitive images.
The learning half consolidates features shipped over the past year. Study Mode walks a student through a problem with guiding questions instead of handing over an answer, and a new Study Hours setting switches it on automatically at chosen times. Responsible homework reminders surface when ChatGPT detects that a student appears to be shortcutting an assignment, and steer them back into Study Mode. Learning visualisations, which covered roughly 70 topics at their March launch, now span more than 300. “We’ve had Study Mode in the field in a number of randomized controlled trials and are seeing positive gains in performance,” Jonas said, declining to share figures.
Parents with linked accounts can set quiet hours when ChatGPT is unavailable, control memory and model-training settings, and receive notifications in a narrow set of high-risk situations. OpenAI is expanding that trigger list to include conversations suggesting an eating disorder, after adding violent-threat violations last month. “We notify parents via text, in-app, and email. All flagged content is reviewed by full-time OpenAI employees before a parent is notified, and we aim to notify parents within an hour of the prompt,” Jonas said.
The age-prediction model itself, detailed by OpenAI in January, reads behavioural and account-level signals: how long an account has existed, typical times of day someone is active, usage patterns over time, and stated age. Adults wrongly swept into the teen experience can restore full access by submitting a selfie to Persona, a third-party identity-verification service. When the system is uncertain, OpenAI says, it defaults to the safer experience. The EU rollout follows in the coming weeks.
Why it matters
The sequencing is the story. ChatGPT arrived in late 2022 and scaled past 900 million weekly users before a purpose-built teen experience existed. A Common Sense Media study last year found that nearly three-quarters of American teenagers use AI for companionship, more than half of them at least a few times a month; a Pew Research Center survey found nearly a third of US teens use AI chatbots daily. The safety architecture is being retrofitted onto an install base that already formed its habits.
That retrofit is happening under visible pressure. OpenAI faces a wrongful-death claim from the family of a 16-year-old, a Florida suit filed in June, and a Massachusetts prosecution citing ChatGPT logs. In September 2025 the FTC opened a Section 6(b) inquiry into companion chatbots and minors, serving orders on seven companies including OpenAI, Alphabet, Meta, Snap, xAI and Character Technologies. California’s SB 243, effective in 2027, will require three-hour break reminders for minors and bar companion chatbots from self-harm and sexually explicit conversations. New York has its own companion-model statute. Much of what launched Tuesday reads like pre-compliance.
Child-safety advocates are not sold. “ChatGPT has made meaningful progress in becoming safer for teens, and it can be valuable for learning and creativity,” said Robbie Torney, senior director of AI programs at Common Sense Media, whose risk assessment nonetheless rated ChatGPT high-risk for teens. “However, we have significant concerns about teens using ChatGPT for companionship and mental health advice at this key developmental stage.” The group’s testing found parent alerts for self-harm content frequently arrived more than 24 hours later — against OpenAI’s stated one-hour target. Founder Jim Steyer has been blunter: “Social AI companions are not safe for kids.”
The deeper precedent is methodological. OpenAI has chosen inference over verification — guessing age from behaviour rather than demanding documents — with biometrics as the appeals process. If that model holds up commercially and legally, every large consumer AI product will copy it, and the cost of proving you are an adult becomes a selfie.
What to watch
Three numbers would settle how much of this matters, and none of them exists yet: how accurately the prediction model identifies under-18 accounts, how many users route around the teen experience, and whether crisis notifications actually reach parents inside the hour. OpenAI has not disclosed accuracy figures from testing. Watch the EU rollout, where age-assurance rules are stricter; watch whether the FTC inquiry produces findings before SB 243 lands in 2027; and watch the distribution picture widen — Google switched on Gemini inside Google Classroom for K-12 schools last week, which means the next cohort of teenagers will meet a chatbot at school before they ever choose one.
“There's no need for a teen to create a new account or change anything; if we predict you're under 18, or you've told us so, this becomes your default experience.”— Lauren Jonas, Head of Youth and Families, OpenAI