For three years, teenagers used the same ChatGPT as everyone else. Same model, same tone, same blind spots. That quietly ended this week. OpenAI launched ChatGPT for Teens, a separate experience for users aged 13 to 17. Buried inside the announcement is something more interesting than a feature list: an admission that general-purpose AI chatbots were never designed with adolescent brains in mind.
The rollout works through OpenAI’s existing age-prediction system. If it estimates a user is under 18, or the user states an age between 13 and 17, the account routes automatically into the teen experience. There’s no manual toggle. No waiting for a parent to flip a switch after something goes wrong. That matters, because 57% of teens already use AI chatbots for schoolwork, often before any parent even knows an account exists.
The real story isn’t safety. It’s friction.
Every previous teen-safety move from an AI company has followed the same playbook: restrict, filter, notify parents, repeat. OpenAI’s version does that too, but it also does something the industry has mostly avoided — it deliberately slows the product down. Responsible homework reminders detect when a teen is trying to shortcut an assignment and redirect them into Study Mode instead of handing over an answer. Study Hours let a teen or parent schedule blocks of time when that step-by-step mode turns on by default.
That’s a strange thing for a company to build into a product it wants people to use more. Every growth team in tech optimizes for reduced friction and increased session time. OpenAI just shipped a feature whose entire job is to make the product slower and more effortful for a segment of its user base. The message underneath it: engagement metrics matter less here than whether a fifteen-year-old actually learns the quadratic formula instead of just copying the output.
Where the real fight is happening
Skim past the study tools and the announcement gets more revealing. OpenAI’s updated under-18 model spec goes further than blocking romantic or sexual roleplay. It instructs the model not to use romantic language, not to encourage emotional dependence, and not to imply it has feelings or consciousness when talking to a minor. That’s a direct response to a problem that has already ended up in court: lawsuits and regulatory scrutiny tied to teens forming attachment-like relationships with chatbots, most notably around Character.AI. Anthropic runs a version of the same debate, and the design philosophies behind Claude and ChatGPT diverge in ways that go beyond raw capability into how emotionally present each model is allowed to be.
OpenAI is essentially telling its own model to stop being likable in a specific way. Not less helpful, not less capable — just less emotionally available. It’s a narrow behavioral carve-out, but it’s the clearest sign yet that AI companies now see “the bot seemed like a friend” as a liability category, not a compliment. The pattern isn’t new; reports of users forming delusional attachments to ChatGPT surfaced well before this launch, and they’re part of why regulators started paying attention in the first place.
Parental controls extend the same logic outward. Parents with linked accounts get Quiet Hours, visibility into select settings, and safety notifications limited to high-risk situations — now expanded to cover eating-disorder-related signals. OpenAI is explicit about keeping that data narrow, sharing only around moments where offline intervention might matter. It’s a deliberate contrast to the surveillance-style parental monitoring apps that scan every message a teen sends.
A curriculum play disguised as a safety launch
The partnership with CodeAI, announced alongside the product, points at where OpenAI actually wants this to go. It’s not just about locking down harmful content — it’s about training a generation of users to treat AI output skeptically by default: understand how the system works, question its answers, catch its mistakes. That’s a materially different pitch than “safe chatbot for kids.” It’s closer to media literacy for a generation that will use AI the way earlier ones used search engines, except with a system that talks back and sounds confident even when it’s wrong.
The showcased student projects reinforce that framing. A Wi-Fi-based search-and-rescue tool built by three Virginia teens, and an audio education platform for blind students that scaled to 200,000 users, both get cited as proof of what teens can build with AI rather than despite the guardrails around it. Whether that’s representative of typical teen usage or a curated highlight reel is a fair question, but it signals the story OpenAI wants told: teens as builders, not just consumers of homework answers.
The part every competitor now has to answer
OpenAI also did something unusual for the industry: it published under-18-specific evaluation results inside its system cards, covering how models perform in sensitive categories like self-harm, eating disorders, violence, and sexual content for younger users specifically. Google, Meta, and Anthropic have all made teen-safety claims. Almost none have published age-segmented model evaluations for the public to check. TechCrunch’s coverage of the launch points out that these age-appropriate protections trace back to OpenAI’s Under-18 Principles, not a one-off feature built for a press cycle.
That’s the detail worth watching past the launch-week coverage. Feature comparisons — Study Mode, Quiet Hours, break reminders — will get replicated by competitors within a quarter. Published, checkable safety benchmarks for minors are harder to fake and harder to walk back. The timing isn’t accidental either: OpenAI is currently facing wrongful death lawsuits and a state attorney general investigation tied to how ChatGPT handled vulnerable users, teen and adult alike. If ChatGPT for Teens becomes the reference point regulators and parents use to evaluate every other AI product teens touch, the study tools won’t be what made the difference. The evaluation data will.
Related: Teens Are Using AI. Schools Still Haven’t Taught Them How to Stay Safe.
