ChatGPT Is Now Guessing Your Age. Here's How That Hits a Work Account.

OpenAI turned something on yesterday that most people are going to read as a kid safety story and scroll right past. It is a kid safety story. It's also going to quietly downgrade some adult accounts, and a few of those accounts belong to businesses around here.

Here's what happened. On Monday OpenAI started rolling out ChatGPT for Teens worldwide, and it should finish inside of two weeks. Anybody who says they're 13 to 17 gets it by default. That part's overdue given everything that's been in the news the last two years.

The piece that matters for the rest of us is the second half of the announcement. ChatGPT now guesses how old you are.

Not from your birthday. From your behavior. OpenAI's own help page lays out the signals it watches: the general topics you talk about, the times of day you use it, how and when the account gets used, and how long the account has existed. And it says plainly that the guess can override the date of birth you typed in when you signed up.

Read that again if your shop runs on one shared login.

Where this actually goes sideways

Picture a small operation with one ChatGPT account everybody uses. Owner's in it at seven in the morning doing quotes. The kid who works the counter after school is in it at four asking about a history paper. Somebody else pokes at it on a Saturday. That account is now a blender of four people's habits, and the system grading it can't tell them apart.

Or say you set up a fresh account last week for a seasonal hire. Brand new, thin history, scattered use at odd hours, nothing consistent about it yet. Nothing in that pattern says "42 year old guy running an HVAC company."

If it flags you, the account still works. You just get the teen version, with reduced sensitive content and some feature limits. Nobody emails you first. You notice because the thing starts answering differently than it did last Tuesday and you can't figure out why.

Getting back out is more of a hassle than you'd think

To pull the teen protections off, you verify your age through a third party company called Persona. Depending on where you are that's a live selfie off your webcam or phone, a photo of a driver's license or passport, or both. OpenAI says it never sees the ID itself and Persona deletes what you upload within seven days.

That's a reasonable setup as far as privacy goes. It's still a thing you now have to do. And it's not something you want to ask an employee to walk through on a Tuesday afternoon so they can finish writing a proposal.

What I'd do about it

Give people their own logins. I've been pushing this for a year for a totally different reason, mostly because shared accounts pile everybody's chat history into one bucket and you can't tell who did what or take anything with you when somebody leaves. Now there's a second reason. One account carrying four people's habits is going to look strange to a system that grades habits.

If you're on ChatGPT Business this is mostly not your problem. Those are managed workspaces, and a seat gets assigned to an actual person. That's a big part of why I move teams onto the business tier once they're past two or three people using it daily. Twenty five bucks a seat and your company's work stops living inside somebody's personal Gmail account.

One more thing worth knowing. If you've got someone under 18 on the payroll using ChatGPT for work, they're on the teen version now whether anybody likes it or not. That's working as intended. Just understand the answers they hand you may be shaped a little differently than the ones you'd get yourself.

None of this is an emergency. It's an hour of housekeeping. I'd rather you spend that hour on a slow afternoon this week than sort it out during the week you're buried.