What happens when a social platform delegates "are you old enough to be here?" to a visual AI system trained on body measurements? Meta is doing exactly that.
The company is running a computer vision system (AI that analyzes images and video) in select countries that estimates whether users are minors by examining their physical characteristics - specifically height and bone structure. According to Meta's announcement, the system is live now, with broader deployment planned.
The regulatory pressure explains the timing. Age verification requirements have tightened significantly - the UK's Online Safety Act, the EU's Digital Services Act, and a growing stack of US state laws all require platforms to do more than accept self-reported birthdates, which anyone can falsify in seconds. Meta has faced specific criticism over Instagram's documented effects on teenage mental health. A passive visual analysis system lets the company claim a technical compliance step without requiring users to upload government IDs.
What "Bone Structure Analysis" Actually Means
This isn't an X-ray scanner. The system works from standard photos and videos users already post. It analyzes markers like facial bone development - jaw shape, brow prominence, how adult features have filled in - alongside proportional body measurements that correlate with age. Adolescent facial structure continues developing through the mid-teens, which gives computer vision something to work with.
Accuracy is the central question. Visual age estimation systems perform reasonably well in controlled test conditions. Social media photos are not controlled conditions. Lighting varies, angles differ, filters distort features. A tall 15-year-old who looks older passes through; a small-framed 21-year-old gets flagged. What happens to flagged accounts? Meta hasn't specified whether the intervention is an ID upload request, a feature restriction, or an account lock. The policy implications of error cases matter as much as average accuracy.
Scanning Everyone to Find Some
The deeper problem: this system doesn't analyze only the users who look young. It analyzes every user to determine whether they might be a minor. That means biometric data processing - data derived from physical characteristics - at social-media scale for billions of adults who are clearly old enough to be on the platform.
Biometric data is treated differently than other personal data under privacy law in many jurisdictions precisely because it's immutable. You can change your password; you cannot change your bone structure. Processing it for a stated child safety purpose doesn't automatically make it compatible with GDPR, the California Privacy Rights Act, or similar frameworks that require explicit justification for biometric collection.
There's also the disparate impact question. AI vision systems trained on datasets that don't represent all body types and ethnicities tend to perform worse on underrepresented groups - a problem documented extensively in facial recognition research since at least 2018. A system that mislabels adult women from certain backgrounds as potential minors at higher rates, or fails to catch younger-looking teenagers from specific demographics, would be both a civil rights problem and a practical failure of the stated goal.
Meta hasn't published training data composition, accuracy rates, or demographic performance breakdowns for this system. The countries where it's currently running haven't been specified. That opacity is a problem regardless of whether the child safety motivation is genuine.