ChatGPT for Teens: What It Blocks, and How It Knows Your Age

OpenAI began rolling out ChatGPT for Teens on August 18 — a restricted mode for users aged 13 to 17, announced on its site and covered the same day by CNBC and TechCrunch. It blocks romantic role-play and other content categories, adds a study-first mode and parental controls, and is switched on automatically: you say you are a teen, or the age-prediction system decides you are. That second path is the part with consequences far beyond OpenAI — and it lands differently in India than anywhere else.

What ChatGPT for Teens actually changes

The teen experience is not a separate app but a state a ChatGPT account enters on Free and paid personal plans, rolling out globally from August 18, per AP reporting. Inside it, ChatGPT declines romantic or sexual role-play and limits depictions of graphic violence, Fortune reports, along with content around viral challenges and extreme beauty ideals. The model is also instructed not to present itself as having feelings for the user — aimed straight at the “AI companion” pattern regulators have zeroed in on. When a conversation suggests risk of self-harm, OpenAI’s help pages say a human review team can assess it and, in serious cases, notify a linked parent.

The school-facing features are the newest part. Study Mode answers homework-style questions with guiding steps rather than finished answers, and parents or teens can schedule “Study Hours” when it is the default, per TechCrunch. A break reminder appears after 90 minutes of use inside a three-hour window, per CNBC. Linked parents can set quiet hours, disable voice mode, memory and image generation, and opt the teen’s conversations out of model training, per OpenAI’s parental-controls documentation — a control adults still exercise through settings, as we detailed in what happens to your chats.

What parents do not get matters as much. They cannot read their teen’s conversations — OpenAI calls that teen privacy; Inc. calls it the catch: the controls govern the container, not the content. The same piece cites survey data that 64% of American 13-to-17-year-olds have used an AI chatbot, 59% of them ChatGPT.

How OpenAI decides who is a teen

Age prediction is the load-bearing wall. OpenAI deployed the system across consumer accounts in January, per CNBC: it weighs the age you stated, how long the account has existed, typical time-of-day usage and other behavioral patterns — and, by OpenAI’s own description in its help pages, defaults to the under-18 experience when unsure. Adults flagged wrongly can restore full access by uploading a government ID and selfie through Persona, a third-party verification service.

The design has two known failure modes. Adults get locked into teen mode — TechRadar documented misfires clustering around newer accounts and late-night usage — and the remedy asks the people angriest about being profiled to hand over identity documents. The inverse also holds: NIST evaluations have found no age-estimation algorithm that performs consistently across demographics, and the bypass toolkit of a false birthdate and a VPN remains available to any determined fifteen-year-old.

A year of lawsuits built this launch

OpenAI committed to age prediction on September 16, 2025 — Sam Altman said then that for teens, safety would come ahead of privacy and freedom, per TechCrunch — three weeks after the family of 16-year-old Adam Raine sued OpenAI over his death, and parental controls shipped that September 29. The pressure has only compounded since. The FTC opened a 6(b) inquiry into companion chatbots and minors in September 2025, naming OpenAI among seven companies. California’s SB 243, the first state law regulating companion chatbots, took effect January 1. Florida’s attorney general sued OpenAI and Altman directly in June. Families of the Tumbler Ridge school-shooting victims filed suits seeking over $1 billion in April, alleging OpenAI ignored a violence flag on the shooter’s account. And the GUARD Act, which would mandate age verification for chatbots nationally, cleared the Senate Judiciary Committee unanimously in April.

Days before launch, Massachusetts prosecutors charged a 17-year-old with murdering his mother and brother and said he had used ChatGPT for “fantasy stories” rehearsing elements of the crime, per CNN; he has pleaded not guilty. No case above has produced a verdict against OpenAI, which denies liability in all of them. But the pattern is the point: every feature in Tuesday’s announcement answers a legal or legislative demand already on file.

How rivals handle users under 18

Against its competitors, OpenAI’s launch is the middle path — neither Character.AI’s exit nor Google’s quiet gating. Character.AI ended open-ended chat for minors entirely in November 2025 after its own wrongful-death litigation, which Google and Character moved to settle in January. Meta paused teens’ access to its AI characters in January and now alerts parents when a teen discusses self-harm with its assistant, in four countries so far. Google offers Gemini to children only through Family Link supervision and blocks image generation for under-18s; Anthropic simply requires Claude users to be 18:

The unflattering precedent for all of them is Instagram. Meta’s Teen Accounts, the closest analogue to retrofitting minor-safety onto a huge existing product, were tested independently last year by the Molly Rose Foundation and Fairplay with academic researchers: fewer than one in five safety tools worked as described, and 64% were rated substantially ineffective, TIME reported. Announced protections and functioning protections are different claims, and nobody outside OpenAI has yet tested which one ChatGPT for Teens is.

The India angle

No market has more riding on this than India. Altman said in February that India has 100 million weekly active ChatGPT users, OpenAI’s second-largest market, and India’s student population is widely described as its largest anywhere — adoption pushed by the ₹399 ChatGPT Go tier being free since last November, a tier we compared in our free-AI-plans breakdown. A mode that changes what ChatGPT says to a student changes the default homework tool for a meaningful slice of Indian classrooms.

But a legal mismatch hides in the age bands. OpenAI’s teen tier treats 13-to-17-year-olds as semi-autonomous users — accounts of their own, parental linking optional. India’s Digital Personal Data Protection Act defines everyone under 18 as a child and requires verifiable parental consent before processing a child’s data at all — with behavioral tracking of children barred outright, and penalties for children’s-data violations running to ₹200 crore. The rules notified in November 2025 phase in: the enforcement machinery switches on this November, full compliance lands in May 2027. An age-prediction system that watches usage patterns to classify minors — the mechanism at the heart of this launch — sits awkwardly beside a statute whose consent must come first, a tension we mapped when we broke down the DPDP Act.

OpenAI knows the collision is coming: it previewed an India-specific Teen Safety Blueprint in February, and Medianama’s analysis catalogued what it left open — how parental consent gets verified, and how always-on age estimation squares with the Act’s data-minimisation duty. The August launch answers neither. Meta, by contrast, built Instagram’s India teen accounts explicitly around the draft DPDP rules, and the government has floated DigiLocker-based age tokens as national infrastructure for exactly this problem. New Delhi hosted an 89-country AI summit in February where the prime minister called for child-safe, family-guided AI; the gap between a 13+ teen tier and an under-18 consent law will not stay unexamined through the DPDP deadlines.

What to watch

Three tests will show whether this is safety engineering or a settlement exhibit. First, independent verification: Instagram’s teen protections looked comprehensive until outside researchers put numbers on them; the same test — do the blocks hold, do the alerts fire in time — has not been run on ChatGPT for Teens. Second, the misclassification ledger: how many adults land in ID verification and how many teens sail through as adults will decide whether age prediction becomes the template the AI tools sector adopts or a cautionary tale the GUARD Act legislates past. Third, India: OpenAI says eating-disorder alerts are coming within weeks, per CNBC, but a DPDP-compliant parental-consent flow for its second-largest market is the harder ship date, and that one carries a statutory deadline. Deliver it before May 2027 and India gets the strictest ChatGPT teen regime anywhere; miss it and 100 million weekly users sit on the wrong side of a ₹200 crore question.