That "Quick" Age Check? It's Quietly Building a File on You

When you verify your age on ChatGPT, the company behind it never actually sees your ID. It never sees your selfie either. A separate service checks your face, then hands over one tiny piece of information, basically just the answer to "is this person over eighteen?"

If you've ever snapped a photo to prove your age

If you've ever snapped a photo to prove your age online, this is already part of your life. And I get why it feels uneasy. Handing your face to an app to unlock something feels like giving away a piece of yourself. But there's a huge difference between an age check that asks one narrow question and one that quietly builds a file on you. Most people don't know that difference even exists. So how does a good age check actually protect you, and how does a bad one betray you?

Let's start with the one that gets it right. When ChatGPT runs age checks through a company called Persona, Persona looks at your ID or your face, figures out your age, and then deletes the raw images within seven days. OpenAI only receives your date of birth or an age estimate, nothing else. That's a principle called data minimization. It means you collect only what you need, then destroy the rest.

Compare that to a different provider called Yoti. Yoti also deletes your selfie after the check, but it holds onto the result for six months. That's not deletion. That's storage. Same photo, very different afterlife. For you, it means two age checks that look identical can treat your data completely differently.

You'd think a machine guessing your age from a

Now, you'd think a machine guessing your age from a photo would be nearly flawless. It's not. According to testing by N.I.S.T., the U.S. standards agency, the best face-age systems are off by about three years on average. And right at the boundary that matters most, sixteen to eighteen, the error margin sits around two and a half years. So some real eighteen-year-olds get wrongly rejected. And some fifteen-year-olds slide right through.

Why so shaky? Lighting, camera angle, makeup, even your expression can throw the estimate off. And if the system was trained mostly on one kind of face, it performs worse on everyone else. That "simple" age check is actually a tangle of engineering problems.

Here's the part almost nobody notices. Before you ever tap "verify," ChatGPT is already guessing your age, from how you type and what you ask. If it thinks you're a teenager, it quietly switches you into a restricted experience. You never opted in. The system just decided.