Facial Recognition Camera: All 8 Wrong IDs Hit Darker Skin

Police in Western Australia scanned more than nine hundred thousand faces in a three-month trial.

Eight people were wrongly flagged

Eight people were wrongly flagged.

Every one of those eight had darker skin.

If you've ever walked through a busy train station or a stadium crowd, this story is about you.

Your face can be scanned there in a fraction of a second, and you'd never know it.

For some people, a wrong match isn't a small glitch. It means a stranger in uniform walks up and says the computer thinks you're someone else.

Here's what happened.

Western Australia Police put mobile camera units in crowded public places and at big events.

The cameras compared every passing face against a

The cameras compared every passing face against a police watch list.

According to reporting from Biometric Update, the system produced two hundred and nine alerts that led to seventy-nine arrests.

Police called it a success, with accuracy near ninety-nine point nine percent.

So what does a number like that actually hide?

Start with the eight.

Police Commissioner Col Blanch said the wrong alerts were linked to darker skin tones, uneven lighting, and weak reference photos.

When every error lands on one group, that's not bad luck. That's a pattern.

It matches what researchers have already measured

And it matches what researchers have already measured.

The Gender Shades study, from M.I.T., tested commercial systems.

It found errors for light-skinned men under one percent.

For darker-skinned women, the error rate was more than a third.

Same technology. Wildly different outcomes depending on who's standing in front of the camera.

The government's own testing points the same way.

According to the National Institute of Standards and Technology, known as N.I.S.T., a study of one hundred eighty-nine algorithms found many misidentified Black and East Asian faces ten to a hundred times more often than white faces.

A tool can be right almost every time on average

So a tool can be right almost every time on average and still fail one community again and again.

For an investigator, that means one headline accuracy number tells you very little.

For everyone else, it means the technology working for most people doesn't mean it works for you.

Now, the missing pieces.

The trial results didn't say who the system missed.

Those are false negatives, people on the watch list who walked right past the camera.

They also didn't break down performance by skin tone or any other group.

Without that, ninety-nine point nine percent is a

Without that, ninety-nine point nine percent is a claim, not proof.

There's a consent question too.

Police say faces of people not on the list were pixelated right away and not saved.

But the scan itself still happened to every person who walked by.

And according to the reporting, police didn't consult the state's privacy watchdog before launching.

Police do have a fair defense.

They say a human officer confirms every match before anything escalates.