249 Arrests, One Question: Will Croydon's Facial Recognition Cases Survive Court?

Two hundred and forty-nine arrests in thirteen months. That's how many people London's Metropolitan Police picked up using live facial recognition cameras on the streets of Croydon. On average, a new arrest every thirty-four minutes the system was running.

Those numbers sound like a success story

Those numbers sound like a success story. And for the people living in Croydon's Fairfield Ward, where crime dropped about twelve percent, maybe it is. But if you've ever walked past a security camera — at a train station, a shopping centre, a football match — this story is about what happens when a machine flags your face, and an officer has seconds to decide what to do about it. The Croydon pilot put live facial recognition on busy streets, scanning crowds in real time against watchlists of wanted individuals. Of those two hundred and forty-nine arrests, a hundred and ninety-three people were charged or cautioned. According to the Met, the system cut the average time to locate a wanted person by more than half compared with their older van-based setups. Impressive speed. But speed creates a question that follows every single one of those cases into a courtroom. When an arrest happens that fast, did anyone build a paper trail strong enough to survive a judge?

To understand why that question matters, you need to know how this technology actually works in the field. Each Croydon operation used a custom watchlist built no more than twenty-four hours before deployment. Once the operation ended, that watchlist was deleted. So the system wasn't scanning faces against some permanent national database. It was a targeted list, assembled fresh each time. When the cameras flagged a potential match, an officer had to review it manually before acting. That's the protocol. A match from the algorithm isn't an identification. It's an investigative lead — a starting point that still needs verification and corroboration. For anyone who's ever been misidentified — at a store, by a stranger on the street — imagine that misidentification triggering a police stop in real time. That's the gap between what the technology produces and what the law requires.

And that gap has real consequences. According to reporting from The Hill, at least eight documented wrongful arrests have been linked to facial recognition across various jurisdictions. In six of those cases, police never checked the person's alibi. Six out of eight. The system gave a lead, and officers treated it like a conclusion.

Different police forces don't even agree on what

Now, different police forces don't even agree on what counts as a strong enough match. Some agencies use a similarity threshold of point-six — meaning the algorithm has to be at least sixty percent confident before it flags someone. Others may set that bar lower. There's no single legal standard that says where the line should be. For a detective building a case, that inconsistency is a problem. For you, it means the threshold that decides whether your face triggers a police stop might depend entirely on which city you're walking through.

The U.K.'s Biometrics Commissioner, William Webster, told The Guardian that the pace of these deployments has outrun the law. Legislation is trying to catch up with what's already happening on the street. And the Equality and Human Rights Commission went further. They described the Met's live facial recognition policy as unlawful, arguing the safeguards fall short and could create a chilling effect on individual rights. That word — chilling effect — means people might avoid public spaces, protests, or gatherings because they know cameras are watching. Even if you've done nothing wrong.

Meanwhile, the documentation challenge is enormous. When facial recognition was used retrospectively — comparing a mugshot to C.C.T.V. footage after a crime — investigators had time. Time to log the comparison, note the confidence score, document who reviewed the match and when. Live deployment compresses all of that. An officer responds to an alert in real time, under pressure. If that stop becomes an arrest, the facial comparison that triggered everything has to hold up under disclosure rules. In the U.S., that means surviving what's called Daubert scrutiny — where a judge decides whether the method behind the evidence is scientifically reliable. Logging who ran the search, what the confidence score was, and what the outcome was for every single alert — that's the kind of documentation experts say agencies need. But when arrests happen every thirty-four minutes, how many departments are actually doing it?