ChatGPT for Teens is a default

On August 18, OpenAI started rolling out a named product it calls ChatGPT for Teens — not a separate kids' app, and not a new signup. If you told the company you are under 18, or its age-prediction model guesses that you are, this becomes the default ChatGPT. The press kit says safeguards. The actual change is classification: OpenAI is now deciding who counts as a minor from usage patterns, then locking that account into a different model spec.

What shipped

The teen experience is for 13–17. Under-13s are still not allowed. OpenAI is rolling it out globally over about two weeks. Linked parents keep the controls that already existed — quiet hours when ChatGPT is unavailable, feature limits, optional safety alerts — and the teen side gets a tighter default without creating a new account. Ads are off for teens in every tier. Parents still cannot read chats. Either side can unlink; the parent is told if the teen does.

Safety filters go further than the adult product: sexual, romantic, or violent roleplay; graphic violence; self-harm depictions; unhealthy beauty standards; risky viral challenges; dangerous activities. Break reminders fire more often and stress that ChatGPT is a model that can wait. Customization stays in wallpaper territory on purpose. Lauren Jonas, OpenAI's head of youth and families, told Engadget the point is to personalize the product without turning it into something that resembles a human relationship.

Prediction in, ID out

OpenAI does not verify age to put someone into teen mode. It uses an age-prediction model on behavioral and account signals — how old the account is, when it is active, usage over time, and the age the user typed at signup. If the system is unsure, it still defaults to the safer experience. Adults who get tagged as teens have to verify with Persona, a third-party ID check, to leave. OpenAI says it does not receive the ID or selfie, only a date of birth or an age prediction. That is an inverted gate: easy to land in the restricted product, paperwork to get out.

The companion problem

The model is instructed not to call itself a friend, claim personal feelings, or imply sentience. That is not a style tweak. It is a response to companionship use — and to lawsuits, including the case over 16-year-old Adam Raine, after which OpenAI said last September it would automatically identify teens and restrict ChatGPT. Linked-parent alerts already covered acute self-harm and violence-related bans. Today they expand toward eating-disorder risk under a broader "serious self-harm" bucket. Flagged cases get a human review by OpenAI staff before a parent is texted, emailed, or pinged in-app; the company aims for within an hour of the prompt, and still does not send the transcript.

Study Hours, not a cheating ban

Study Mode — Socratic hints instead of finished essays — does not become the default. Study Hours can turn it on at set times, chosen by the teen or a linked parent. Even outside those hours, OpenAI says ChatGPT will try to catch homework-skipping and redirect into Study Mode. Interactive math and science visuals now cover 300-plus topics. OpenAI cites its own number that nearly 9 in 10 teens on ChatGPT in a given week use it for learning, information, skill-building, or productivity. Treat that as a platform stat, not an independent audit. The classroom fight is older than this launch; see Gen Z's AI pushback.

What the new default does not close

Age prediction is not proof of age. Teens who want the adult product will try to look like adults; adults who get stuck will hit Persona. Unlink still kills parental alerts. Quiet hours and Study Hours only bind a linked, cooperative household. TechCrunch's timing point is fair: ChatGPT shipped in late 2022 and grew to a reported ~900 million weekly users before a teen-specific default existed. Filters also fail in the usual ways — jailbreaks, alt accounts, a parent's phone. Until independent testers try to break this, "safe for teens" is a product claim.

What directory readers should watch

If you evaluate consumer chat products, the checklist is now age gating, not only content policy: how the vendor infers age, who can appeal, whether parents see transcripts, and whether teens can walk out of the restricted mode alone. ChatGPTsits in Language Models; the educator surface is listed as ChatGPT for Teachers under Teacher Tools. Related classroom products live in Education. Our Ethics page is the standing filter for youth-safety claims that do not say how classification actually works.

The point

OpenAI now applies teen restrictions by default to anyone it believes is under 18, and adults who are flagged have to verify their age to get out of them. The companion limits and parent alerts respond to the lawsuits against OpenAI, but the age-prediction model decides who actually gets those protections. Watch how often it misclassifies adults, how often teens unlink parent accounts, and whether Study Hours changes how students do homework or just pushes them to unmonitored accounts.