Facial Recognition Security: 1.7 Million London Faces, 44 Arrests
Quick answer
Does UK police facial recognition only target wanted criminals?
The UK says police should use live facial recognition only to look for wanted criminals, but that describes the watchlist, not the cameras. A camera has to scan every passing face to find those few people. What happens to the other faces, and how wrong matches are corrected, is not clearly answered.
Picture walking down a busy London street to buy milk. A camera on a van scans your face in a split second, and you never know. That is facial recognition (software that compares faces to a list of people police want to find), and the UK says facial recognition should only be used to hunt wanted criminals.
Facial recognition limited to wanted criminals sounds reassuring, but a limit on who police look for says nothing about what happens to everyone else's face or how mistakes get fixed.
Facial recognition limited to wanted criminals sounds reassuring, but a limit on who police look for says nothing about what happens to everyone else's face or how mistakes get fixed.
First, the straight answer to the 11pm question. Are you in danger? Probably not in any dramatic way. But you deserve clear answers, and right now the headline promise doesn't give them. Let's walk through it.
What Is UK Police Facial Recognition Being Limited To?
The Telegraph reports that Home Secretary Shabana Mahmood has told police to use facial recognition cameras to find wanted criminals. I could not pull up that exact article, so I'm leaning on wider coverage of the same announcement. As Biometric Update described it, the plan includes a new watchdog (an independent body that checks the rules are followed) with the power to seek injunctions (court orders) to stop facial recognition technology it judges unsafe.
How live facial recognition works on the street
Live facial recognition means cameras scan faces as people walk past. The software first uses face detection (finding a face in the video frame), then each face image is turned into a match score (a number for how alike two faces are) and compared with a watchlist (a database of the people police are looking for). Under the new approach, use should be intelligence-led (based on information that wanted people may be in the area) and proportionate (not bigger than the problem). This article is part of a series, start with Facial Recognition False Arrest 9 Wrong Alerts 1 38m Faces.
That is a fair goal. Nobody sensible wants a violent fugitive walking free past a camera that could have spotted them. But here is where it gets interesting.
1.7 million
face scans in London in the first four months of 2026, leading to 44 arrests
Source: research summary of London deployment figures
Does Facial Recognition Only Hunting Wanted Criminals Protect Everyone Else's Face?
No, not by itself. The camera has to look at every face to find the few it wants. Do the math on the London figures: 1.7 million scans and 44 arrests works out to roughly 38,600 faces checked for every arrest. Almost everyone scanned is just a person going about their day.
So "wanted criminals only" describes the target. It does not describe what happens to the other 38,599 faces. Are they deleted at once? Stored? Who can get access to them? Those are the questions the headline skips. Previously in this series: Child Identity Theft 3 In 4 Young Victims Know The Thief.
What happens when police facial recognition gets it wrong?
Mistakes are the other gap. A false match means the system flags the wrong person. Testing cited in the research found false positive rates up to 100 times higher for Black and Asian faces than for white male faces, with weaker accuracy for women and younger people. The Home Office's own tests of a new algorithm reportedly found high accuracy and no bias. Those two claims sit awkwardly together, and a reader is right to ask which one to trust.
A "wanted criminals only" rule tells you who the police are looking for, but tells you nothing about who else gets flagged, how errors are corrected, or how innocent people discover they were mismatched.
Expert research summary, drawing on Statewatch and related coverage
Why UK Police Facial Recognition Rules Still Leave Security Gaps
| What the limit covers | What it does not clearly answer |
|---|---|
| Who police say they are looking for | What happens to everyone else's face after the scan |
| Use that is intelligence-led and proportionate | How an innocent person learns they were wrongly matched |
| A new watchdog with power to seek injunctions | Whether error rates are published for every group |
| Live facial recognition cameras on the street | Retrospective searches (checking old images against police databases), which have also produced wrong identifications |
To be fair, the other side has a point. A narrower watchlist and a watchdog with teeth are real improvements. Roughly 80% of people surveyed in England and Wales said they were comfortable with police using facial recognition to search watchlists, according to the research. Public support is real.
Why facial recognition limits need more than a target list
- Scale - a camera must scan thousands of ordinary people to find one wanted person.
- Errors - a false match lands on a real person, and the uneven error rates make it worse.
- Paper trail - an audit trail (the record of who looked at what, and when) is how anyone checks the rules are followed.
One useful thing to watch for: identity and image claims
If you have ever wondered whether a face in an image really matches the identity it claims, that is exactly the identity verification question facial recognition technology exists to answer. The honest version of the answer always comes with a number and a margin for error. So when anyone says a system is accurate, ask: accurate for whom, and how often is it wrong? If they cannot say, treat the claim as marketing.
UK police facial recognition limited to wanted criminals is a fair goal, but a target list is not a promise about ordinary faces or mistakes.
Recap: Limiting cameras to wanted people sounds safe, yet it leaves open what happens to everyone else's face. It also leaves open how errors get fixed. Up next: Facial Recognition 1 7 Million London Faces 44 Arrests.
So here is my question for you, and I'd honestly love to hear your answer. Would you be comfortable with a police camera checking your face for a wanted person? What would you need to know first? For me, it's simple: tell me how long my face is kept, and who do I call when the camera is wrong?
Facial Recognition Technology: Frequently Asked Questions
Is facial recognition used by police only for wanted criminals?
The UK guidance says it should be. That describes who officers are meant to look for, not what happens to the many other faces a camera scans. Coverage of the plan also describes a watchdog and testing requirements. Whether the limit is followed in practice depends on checks, records and public reporting, which is why people keep asking for clear answers about stored images and mistakes.
What does uk police facial recognition do with faces that do not match?
The sources I found do not spell this out, so I won't guess. That gap is the point of this article. The reasonable thing to ask is how long non-matching faces are kept, who can get access to them, what security protects them, and whether any record of the scan exists. A clear public answer should exist for each of those.
Can facial recognition make mistakes?
Yes. Any system can flag the wrong person. Research cited in the sources found false positive rates up to 100 times higher for Black and Asian faces than for white male faces. The Home Office's own tests of a new algorithm reportedly found no bias, so the evidence is contested. Either way, a fair system needs a way for wrongly matched people to find out and challenge it.
What is retrospective facial recognition?
It means searching previously captured images, such as a still from a camera, against police databases, instead of scanning faces live. The research summary notes this has also produced wrong identifications. It says the new guidance focuses on live use and leaves these offline searches without equivalent safeguards, which matters because errors there can persist unnoticed.
Ready for forensic-grade facial comparison?
Full forensic reports with detailed similarity scoring. Results in seconds.
Run My First SearchMore News
AI Voice Cloning: Fake Call Takes Rs 1.2 Lakh From Florist
A Kolkata shop owner heard a voice she'd known for years and sent money fast. It wasn't that person. Here's the simple pause that beats AI voice cloning.
privacyTikTok's Response: Nearly 50 Nude Deepfake Apps, 'Not Ads'
A report says nearly 50 apps that make sexual deepfakes were pushed through TikTok ads. TikTok says the examples weren't ads. Either way, a familiar feed is not a safety guarantee.
facial-recognitionFacial Recognition in Stores: False Match Ejects UK Shopper
A shopper was thrown out of a UK store after a face-match alert said she'd stolen meat. She says she hadn't been there in months. Here's why an alert isn't proof, and what to say if it happens to you.