Instagram's AI now guesses your age everywhere. Meta won't say how often it's wrong

If you lied about your birthday when you signed up for Instagram, Meta would like you to know it has been reading your posts.

On November 9 the company announced that the AI it uses to find teenagers on Instagram is going global. "We use AI technology to find people we suspect to be teens, even if they use an adult birthday," Meta wrote. Accounts it flags get moved into Teen Accounts, which "automatically limit who can contact them, the content they see, and their time on our apps." Meta says the expansion means "millions more teens across the world who use Instagram will receive these built-in protections."

Meta says it was "encouraged by the results" of the earlier rollout. It did not share them.

What the machine reads

The November post is thin on mechanics. For those you have to go back to Meta's May announcement, when the system spread to "27 countries in the EU and Brazil" after launching in 2025 in the US, UK, Canada and Australia. There, Meta said its AI looks at "entire profiles for contextual clues," with birthday celebrations and mentions of school grades as its examples, drawn from "posts, comments, bios, and captions." Think of a friend's happy-sixteenth comment, or a caption moaning about a Year 10 exam.

In some countries Meta also runs what it calls visual analysis on photos and videos. The May post said "this is not facial recognition," and a correction added on July 31 went further: "our AI does not use biometric data to identify whether a person is underage. It does not create facial fingerprints or identify who is in a photo." Meta has not said which countries get the visual analysis.

MediaNama, the Indian tech policy site, says Meta hasn't disclosed which signals it uses. That overstates it, since the May post names some. But MediaNama is right on the bigger point. There is no full list, no weighting and no accuracy figure. Notebookcheck put it plainly: "Meta has not provided a global accuracy figure or false-positive rate for the expanded system."

The adults caught in the net

That missing number matters because a classifier like this is wrong in both directions. Some teenagers will slip past it. Some adults will almost certainly be told they look fifteen. The November announcement does not spell out the route back for an adult wrongly moved into a Teen Account. The routes Meta has described for related cases run through documents. If an account is deactivated as suspected underage, "the account holder will need to provide proof of age through our age verification process." Anyone changing a birthday from under 18 to over 18 has to "verify their age using an ID or Yoti's facial age estimation tools to complete the change."

So the way out of one AI's guess about you is a government ID, or a second company's AI guessing your age from your face. Meta's November post complains that some age laws "require invasive age checks for everyone." Its own fallback looks a lot like one.

Apple as the bouncer

In Australia, which banned social media for under-16s in December 2025 according to MediaNama, Meta is also using Apple's age-range signal. A parent sets up the phone, the App Store assigns an age range, and "each time the teen downloads an app, Apple passes that age range to the developer." Meta says it is using the signal voluntarily, "as long as collecting it and sharing it remains optional". TNW and Notebookcheck both report the Apple signal is Australia-only for now.

Meta's preferred future is one where the phone does the checking for everyone. "One verified signal from people's devices, shared consistently with all apps, is far more effective and privacy-preserving than asking every platform to verify age individually," it wrote. Then comes the interesting part. Meta says its own estimates "can be used as a backstop to help correct mistakes and feed them back to the app store for correction across all the apps a teen uses."

Read that one again. A guess made by Meta's model, from a teenager's captions and comments, could travel upstream to the app store and change how every other app on the phone treats that person.

The timing is not subtle either. Meta's post cites its agreement with "52 bipartisan attorneys general" across US states, territories and the District of Columbia, and TNW reports that a New Mexico jury found Meta liable and ordered it to pay $375 million. Threads gets similar technology "later this year." Facebook is still being rolled out. US parents will start seeing an in-feed video explaining it this month.

Every big platform is now being pushed to know how old its users are. Meta's answer is to infer it from what they post, worldwide, with the error rate kept to itself.

Sources: Meta newsroom, Meta newsroom, May 2026, The Next Web, MediaNama, Notebookcheck.