OpenAI began a global rollout of a separate ChatGPT for Teens mode on August 18, automatically switching in any account its own systems predict belongs to a 13-to-17-year-old — based on account signals and behavior patterns, not just a self-reported birthdate — or that a user identifies as being in that age range at signup. The company says the rollout will finish within about two weeks. OpenAI has not disclosed an accuracy rate for the age-prediction system itself.
The restricted mode adds evaluations for self-harm, eating disorders, violence, age-restricted goods, and sexual content, and stops ChatGPT from calling itself a friend or suggesting it has feelings or sentience — a relational-boundary rule aimed at the kind of prolonged, personal exchanges now central to multiple lawsuits against the company. Parents who link accounts get Quiet Hours they control and a new Study Mode default that guides students through problems rather than completing them; the company says flagged self-harm or violent-threat content is reviewed by a full-time OpenAI employee before a parent is notified, targeting under an hour.
What changes in the teen mode
| Standard ChatGPT | ChatGPT for Teens | |
|---|---|---|
| Self-harm / eating-disorder content | Governed by general content policy | Evaluated and restricted; high-risk flags trigger a parent alert |
| Age check | Self-reported birthdate only | Self-reported age plus behavioral age-prediction |
| Relational language | Can describe itself in personal terms | Restricted from claiming feelings, sentience, or friendship |
| Parental account controls | None | Quiet Hours, notifications, linked account |
The launch lands directly inside that pressure. OpenAI is defending itself against a wrongful-death lawsuit — Raine v. OpenAI — filed in San Francisco County Superior Court in August 2025 by the parents of Adam Raine, a 16-year-old who died by suicide in April 2025 after months of conversations with ChatGPT. An amended complaint filed that October reframes the case from reckless indifference to intentional misconduct, and alleges OpenAI weakened its own self-harm guidance in February 2025 — after which, per TIME's reporting on the filing, the share of Raine's daily conversations involving self-harm content rose roughly tenfold, from about 1.6% in January to about 17% in April, the month he died.
OpenAI did not respond to TIME's request for comment on the amended complaint, and the company has not issued a public statement addressing the safeguard-weakening allegation directly. In launching ChatGPT for Teens, it frames the new restrictions as an extension of safety work already underway rather than a response to any specific case.
“Won't call itself a friend, suggest it has personal feelings, or suggest sentience.” — OpenAI, describing ChatGPT for Teens' relational-boundary restrictions
The launch also arrived alongside organized opposition. Student activists under banners including Sunrise Movement and QuitGPT occupied OpenAI's Washington, D.C. office the same week, part of a campaign SFist reported spanning more than 20 states; police arrested 13 demonstrators at the D.C. office. (A Pew survey cited in the same reporting found nearly a third of US teens already use an AI chatbot daily, with ChatGPT the most-used one -- the scale that makes both the safety features and the lawsuits matter beyond any single case.)
- The age-prediction system accurately identifies teen accounts.
- OpenAI notifies parents of high-risk flags within about an hour.
- OpenAI weakened self-harm safeguards in the months before Adam Raine's death.
OpenAI is not the only consumer AI company answering this pattern with a fenced-off teen product. Character.AI, the AI-companion app, rolled out its own under-18 guardrails after fighting comparable child-safety lawsuits over conversations that turned personal in the same way regulators and plaintiffs' lawyers now describe in the ChatGPT cases. The through-line across both companies is the same: age-gating and content restrictions arrive after the lawsuits, built around exactly the harms the complaints describe, rather than ahead of them.
That sequencing is also why the legal theory shift in the amended Raine complaint matters beyond one family's case. Moving from reckless indifference to intentional misconduct raises what the plaintiffs have to prove — that OpenAI knew about the risk its own February 2025 policy change created and pressed ahead anyway — but it also raises what is at stake if they succeed: a finding of intentional misconduct is the harder standard a product-liability defendant works hardest to avoid, precisely because it does not treat the harm as a side effect of a general-purpose product.
None of the specific numbers in this launch — the age-prediction system's accuracy, the real-world notification-time average, the outcome of the safeguard-weakening claim — are independently verified yet. What's confirmed is the shape of the response: a separate product tier, arriving roughly a year into litigation the company has not yet had to answer at trial.
- OpenAI launched a separate ChatGPT for Teens mode globally on August 18, 2026.
- It auto-enrolls accounts an age-prediction system flags as belonging to 13-to-17-year-olds.
- New restrictions cover self-harm, eating-disorder, and sexual content, plus parental notifications within about an hour.
- It launched as OpenAI defends an amended wrongful-death lawsuit alleging weakened self-harm safeguards.
- The age-prediction system's accuracy and the lawsuit's core safeguard claim are both still unverified.