Social Media Ban: Why AI Can’t Reliably Tell If You’re 15 or 16: The Limits of Age Estimation

As the Australian Government moves closer to enforcing age restrictions on social media through the proposed Social Media Ban, one of the biggest unanswered questions remains: how do we actually verify someone’s age online?

The Government’s Age Assurance Technology Trial (AATT) set out to test exactly that. Twelve systems from around the world, ranging from facial recognition AI to hand-movement analysis and government ID verification, were submitted as options for the verification of ages.

The key finding? If you don’t have a government-issued ID, there’s no foolproof way to prove your age with certainty.

Facial Estimation Isn’t Precision Tech

Yoti, one of the most prominent facial age estimation providers, claims a mean absolute error (MAE) of around 1.2 to 1.3 years for people aged 6 to 17. That sounds impressive on paper, but in practice it means someone who is 15 years and 10 months old might be estimated as 16.9, or vice versa.

Even the best systems rely on what’s known as an “age buffer.” For example, to reliably catch under-13s, systems apply a 6-year buffer – flagging anyone who looks under 19. This works well for broad age gating (under 13, under 18), but fails miserably when the cutoff is precise, like 16 years old.

Hand-Waving AI? Still Not Precise

Needemand’s BorderAge system uses hand movement analysis via webcam to estimate age. It’s clever, privacy-preserving, and claims over 99% accuracy at identifying whether someone is over or under 18. But again, it’s not designed to tell if you’re exactly 16. Like others, it gives binary outputs: over or under a threshold, with no DOB-level granularity.

Only Government ID Can Prove Age

PRIVO and Persona, two of the most robust platforms in the trial, rely on traditional identity verification methods. They use government-issued ID, sometimes paired with selfies or NFC chip reading, to validate exact date of birth. These are the only systems capable of reliably distinguishing between someone who is 15 years and 10 months old versus someone who is 16 and two days.

But here’s the catch: most minors don’t have a government-issued ID. And without one, these systems either fall back on facial estimation or route the process through a parent or guardian.

So What Now?

The AATT’s findings confirm what many privacy and security experts have been saying for years: AI can help with broad age detection, but it can’t replace proper identity checks.

As policymakers consider how to implement age restrictions for online services, they’ll have to reckon with a tough reality: if you want certainty, you need a verified ID. Anything less is, at best, a smart guess.

For now, AI can help enforce the spirit of the law – but not the letter. And for age gates where legal liability is on the line, a guess just isn’t good enough

Why does this matter?

Well the Government is saying to kids – you have to wait until you’re 16 to get Social Media. Trust me, those kids are going to want access on their birthday, not a day later.

So these systems need to be accurate – not just close.