Facial Recognition: 1.7 Million London Faces, 44 Arrests

Police in London scanned the faces of one point seven million people in the first four months of this year. Those scans led to forty-four arrests. Do the math, and it's roughly thirty-eight thousand people checked for every single arrest.

If you've ever walked down a busy London street, or

If you've ever walked down a busy London street, or any busy street, you may have been one of those faces. You wouldn't have been told. You wouldn't have been asked.

Now the British government has announced new rules. According to the Home Office, police may use live facial recognition only when it's intelligence-led and proportionate. In plain terms, officers must believe wanted people could be in the area. Home Secretary Shabana Mahmood announced the framework in early November. It also creates a new watchdog. So here's the question. Does a rule about who police are looking for protect the people who aren't being looked for?

Start with those numbers again. Forty-four arrests is not nothing. But the other side of that ratio is a crowd of ordinary people, each scanned in a moment, each compared against a list. When the scale is that large, the target of the search matters less than how many people the camera touches.

That brings up accuracy

That brings up accuracy. According to testing cited in the research, false positive rates ran up to a hundred times higher for Black and Asian faces than for white male faces. A false positive means the system says you match someone on the list when you don't. The same testing found weaker accuracy for women and younger people too.

A hundred times. Picture two people walking past the same camera. One is far more likely to be wrongly stopped, just because of their face.

And a narrower mission doesn't fix that. If the watchlists lean toward certain communities, the bias can compound.

Then there's a second kind of use

Then there's a second kind of use. It's called retrospective facial recognition. That means police take a photo from a past event, like a store camera image, and search it against their databases. According to the article, that method has also produced wrong identifications. The new guidance focuses on live use. The offline searches don't get the same safeguards, so errors there can sit unchecked.

For investigators, the lesson is clear. Whether you compare faces by hand or with software, the question isn't intent. It's how often the tool points at the wrong person. For everyone else, it means a match isn't a fact. It's a guess with a score attached.

Public opinion adds another layer. According to a survey across England and Wales, almost eighty percent of people feel comfortable with police using Utah Age Verification Adult Sites Can Keep Id And Selfies to search watchlists. And the Home Office says its independent tests of a new algorithm found high accuracy and no bias. That claim sits uneasily next to years of research showing demographic gaps. People may be comfortable because they assume the tool is accurate. Is that a promise anyone has actually made?

To be fair, the rules do matter

To be fair, the rules do matter. A narrower purpose means smaller watchlists. Fewer arbitrary choices about where to point the cameras. And the new watchdog can seek injunctions to stop a technology if it's judged unsafe. That's a real circuit-breaker if real-world errors drift from what the tests claimed.

The new rule tells you who police are hunting. It doesn't tell you who else gets flagged, how mistakes get fixed, or how an innocent person ever finds out they were mismatched.

Most places, including the U.K., haven't required police to disclose any of that.