Open ChatGPT this week and, without any announcement in the chat window, it may already have formed an opinion about how old you are. From 18 August, OpenAI began rolling out “age prediction” on its consumer plans: a system that guesses whether your account belongs to someone under 18 and, if it decides you’re a minor, quietly switches you into a restricted version called ChatGPT for Teens. You are not asked. If the guess lands on “teenager,” the guardrails go up by default.
Answer first, because the mechanism matters more than the alarm: the guess is behavioural, and it is admittedly imperfect. By OpenAI’s own account the system reads “general topics you talk about, the times of day you use ChatGPT, how and when your account is used, and how long your account has existed.” And the way to make it stop guessing is not a toggle. It is to prove your age to a third-party verifier called Persona, with a live selfie, a government ID, or both. The choice on offer isn’t whether to be identified. It’s how.
None of this arrives from nowhere, and we’ll be fair about why in a moment. But a change that infers a protected characteristic from the content of your conversations, applies real restrictions on the strength of a guess, and offers identity verification as the only exit is worth reading slowly — especially for the adults who will be misclassified, because OpenAI says plainly that some will be.
What OpenAI actually switched on
The launch has two parts. The visible one is ChatGPT for Teens, announced on 18 August: a version with study-focused features and stronger safety defaults for under-18s. The consequential one is age prediction, the system that decides who gets dropped into it. In OpenAI’s words, “If our system estimates someone is under 18 or they state their age is between 13 and 17, they are automatically placed into ChatGPT for Teens.” It is rolling out globally, with the EU following “in the coming weeks” to fit regional rules.
What does the teen experience actually change? The safeguards are broad. ChatGPT for Teens reduces sensitive content — OpenAI lists graphic violence or gore, “viral challenges that could push risky or harmful behavior,” sexual, romantic or violent roleplay, and content that promotes extreme beauty standards or unhealthy dieting. Under its updated under-18 model rules, the assistant “should not use romantic language, encourage emotional dependence, or imply that it has feelings or consciousness.” There are break reminders, product cues that keep identifying the thing as AI, and, for accounts linked to a parent, controls like Quiet Hours. OpenAI also says it will not show ads in a teen account — a notable line in the same week it announced that ChatGPT Ads is expanding across Europe for everyone else.
How it decides you’re a teenager
Here is the part that should give a thoughtful adult pause. Age prediction does not work from a document or a birthday you confirm. It works from behaviour. OpenAI’s help page describes it as looking at “different signals linked to your account” — the topics you raise, when you tend to log in, the pattern of your usage, the age of the account itself — and then inferring whether you’re over or under 18.
Crucially, this overrides what you told it. “Age prediction is an extra safety check,” the page reads. “It can identify an account as eligible for the ChatGPT for Teens experience even if you already gave your date of birth when you created your account.” So the date of birth you entered is not the last word; a statistical read of your habits can be. And OpenAI is refreshingly blunt about the reliability of that read: “No system is perfect. Sometimes ChatGPT may get it wrong.”
That candour is welcome, and it is also the whole problem. A system that infers a legal category from “the general topics you talk about” is guessing, and a guess about age from conversation topics is exactly the kind of proxy that misfires — the adult who uses plain language and logs on after school hours, the teenager who writes like a lawyer. When the guess is wrong, it doesn’t fail quietly in a log somewhere. It changes the product you’re using.
The opt-out is a selfie or a passport
Say the guess lands wrong and you, a grown adult, find yourself in the teen experience. OpenAI has built a way out, and to its credit it’s documented and not buried: go to Settings, choose “Verify age,” and complete a check with Persona. Depending on your country, Persona “may ask for one or both of these: a live selfie” — a real-time photo of your face — and “a government ID,” such as a driver’s licence or passport.
The privacy engineering here is genuinely more careful than the norm. OpenAI says it “does not receive the ID itself”; Persona handles the document and “deletes your uploaded ID or selfie within 7 days.” OpenAI only learns the outcome — that you verified as 18-plus, plus age-related information such as your date of birth. If you don’t want the hassle, you don’t have to verify at all; you can keep using ChatGPT with the teen protections on. That is a real set of mitigations and we won’t pretend otherwise.
The system offers two ways to establish that you’re an adult: let it read your habits, or show it your face. There is no third door marked “none of your business.”
But look at the structure of the choice, because it’s the same shape whichever way you turn. You can be profiled — scored for age from the content of your chats — or you can hand a selfie and a government ID to a company you’d never heard of until this settings page. The one option that doesn’t exist is to be left alone: to use a general-purpose writing tool without either a behavioural age-score or an identity check attached to your account. OpenAI’s own wording makes the trade explicit: “If you do not want OpenAI to predict your age, you can verify your age with Persona.” Don’t want to be guessed about? Then prove who you are.
The steel-man: this did not come from nowhere
Now the fair part, and it’s substantial. OpenAI is not doing this for fun, and it is not the only company being pushed here. The launch follows a run of lawsuits and public cases alleging that ChatGPT contributed to serious harm to young users, including self-harm; OpenAI has denied wrongdoing in some of those cases, but the pressure — legal, regulatory and moral — is real and it is not frivolous. A chatbot that will talk to a distressed 15-year-old at 3am is a genuinely different product from a search box, and pretending otherwise would be its own kind of dishonesty.
The company also cites a defensible premise: it says the overwhelming majority of teenagers use ChatGPT for ordinary things — homework, learning, building. Age-appropriate defaults for minors are a reasonable thing to want, regulators increasingly require some form of age assurance (Italy already forces verification within 60 days of being asked), and OpenAI has clearly spent effort making the verification path privacy-preserving. Concede all of it. A world where a company shrugged and served every user the identical unguarded model would not obviously be better.
Why “we might get it wrong” is doing a lot of work
Grant every one of those points and the objection doesn’t vanish; it just gets specific. The issue is not that OpenAI built protections for teenagers. It is that the gate to those protections is a guess, the guess is on by default, and the cost of the guess being wrong is borne by the person it’s wrong about.
Think about who the false positives are. Age estimation is least reliable exactly at the boundary that matters — the 18 line — and its errors are not evenly distributed across everyone; independent testing of these systems keeps finding they’re worse for some groups than others, a point we unpack in our companion explainer on why age estimation is so hard to get right. So the adult who gets dropped into a restricted experience, then asked to upload a passport to climb back out, is not a random unlucky soul. They’re a predictable output of a system that trades accuracy for coverage. When that happens, the restrictions we’ve criticised elsewhere as over-refusal arrive not because you asked for something dodgy, but because a classifier mislabelled you.
There’s a quieter cost too. Inferring your age from “the general topics you talk about” is, unavoidably, profiling the content of your conversations to derive a fact about you — a small expansion of the surveillance that already underwrites why every AI wants your data, now wearing a child-safety badge. And normalising a face scan or an ID upload as the price of the un-nerfed product is the sort of thing that feels reasonable once and becomes the industry default by the third time, in the same way defaults quietly reset in your workplace tools.
What to actually do
The practical advice is undramatic:
- Know that the guess exists. If ChatGPT suddenly feels more restrictive, you may have been placed in the teen experience by prediction, not by anything you did.
- Decide before you verify. You do not have to hand over ID to keep using ChatGPT — you only need to if you want the teen protections removed. Weigh whether the unrestricted version is worth a selfie and a document to a third party.
- If you do verify, note the boundaries. OpenAI says it never receives the ID and Persona deletes it within seven days — commitments worth holding them to.
- Watch the defaults. The interesting question isn’t this launch; it’s whether behavioural age-scoring becomes a permanent, invisible layer on every account, and whether “verify to opt out” spreads from teen safety to everything else.
OpenAI has done the hard, unglamorous safety work more thoughtfully than most, and the goal — keeping the sharpest edges of a persuasive chatbot away from children — is one worth taking seriously. The fixable fault is the design that makes its own inaccuracy your problem. Age assurance built well would treat a wrong guess as a bug the company eats, not a burden the user carries with a passport in hand. Until then, the honest summary is the one OpenAI wrote itself: sometimes it will get it wrong — and when it does, proving otherwise is on you.
Originally published at theaidownside.com — evidence-first reporting on the costs and trade-offs behind AI products.
