500,000 caras escaneadas, 0 arrestos: el reconocimiento facial que fracasó en Gran Bretaña

Half a million faces. One hundred police hours. Three hundred and twenty thousand pounds. And exactly zero arrests — plus one false positive that almost ruined an innocent person's life.

That's the scorecard of the British Transport Police's six-month live facial recognition trial across some of London's busiest railway stations, revealed this week by Liberty Investigates and reported by The Guardian. The document came out through a freedom of information request, and the numbers are brutal: £320,786 spent across 18 deployments between February and July, producing a single watchlist alert — which was wrong.

Read that again. The entire surveillance apparatus scanned over half a million faces to catch one person, and it didn't even catch the right one.

What actually happened

The trial ran across major transport hubs including King's Cross, using live facial recognition — cameras that compare passing faces against a police watchlist in real time rather than after the fact. The stated goal was catching wanted offenders and people breaching court orders.

More than half of all police forces in England and Wales have now deployed this technology somewhere on British streets, according to Liberty Investigates. London is the most aggressive: the Met has announced fixed cameras for the West End, and mayor Sadiq Khan confirmed Oxford Street will be next. The trial is being extended four more months, now including Underground stations.

BTP claims the extension has already produced "three positive confirmed alerts" of people complying with sexual harm prevention orders. They also insist officers made "a number of associated arrests" during the pilot — assault, theft, weapon possession — but that those "did not result directly from an LFR alert" and therefore aren't counted in the performance data.

That caveat is doing a lot of heavy lifting. It's the difference between "our system caught nobody" and "our system is exempt from counting."

Why the false positive matters more than the misses

Privacy campaigners will point at the zero arrests. The people who should be paying attention are the ones the one false positive hit.

A false positive is not an inconvenience. It means a human being was publicly identified as a criminal suspect in a busy train station, on camera, surrounded by commuters — and was wrong. The Guardian's own archive of this technology's UK failures is a roll call: an innocent man arrested for a burglary 100 miles away from the crime, a false shoplifting accusation that forced a Sainsbury's store to pause its scanning, and shoppers who struggled for years to clear their names after being flagged as shoplifters in front of their families.

Nobody is harmed by a system that never alerts. Everyone is harmed by a system that alerts and is wrong.

The uncomfortable comparison

Fraser Sampson, a former UK biometrics and surveillance camera commissioner — who now sits on the board of Facewatch, a company selling exactly this kind of retail scanning system — offered the sharpest critique in the whole story:

"Success in a shop means no such people coming in. Success for the police trying to catch people means people being caught. In that respect the trial doesn't appear to have been very fruitful."

He also flagged what nobody outside the industry wants to admit: the watchlist is the real system. The cameras are just the matcher. If you're not on the list, live facial recognition doesn't "see" you in any meaningful way — and it doesn't retain your image the way ordinary CCTV does.

That means expanding the deployment doesn't automatically produce better results. It requires a bigger, better-curated list. And Sampson's warning is that these cameras process "special category personal data" — a legal category with extra protection under UK data law — so proportionality has to be demonstrated, not assumed.

The regulatory gap nobody has closed

The UK Information Commissioner's Office has recently reported that oversight boards are lagging behind the technology. Parliament's Joint Committee on Human Rights called live facial recognition a "particularly clear example of risk" and demanded a law addressing AI threats to human rights.

Bell Ribeiro-Addy, Labour MP for Clapham and Brixton Hill, put it plainly: these systems are "a serious invasion of our privacy" and the government "should be suspending their rollout until there are stronger safeguards."

The technology isn't useless. The accuracy, when properly tested, is genuinely high. But accuracy is a measure of matching faces to a list, not of being right about a person. The trial just demonstrated the gap between the two with real money and a real innocent person's afternoon.

What this means if you're in Latin America

You'll see this pitch arrive. Some agency or private security firm will offer "advanced AI identification" for your mall, your transit system, your building. The sales deck will quote accuracy percentages and never quote the false-positive rate or the cost-per-actual-identification.

Do the math on this one: £320,786 for one incorrect alert. That's roughly £320,000 per false positive. Any vendor who can't show you their watchlist curation process, their data retention policy, and their human-review step is selling you the same lottery ticket the Met just bought 500,000 times.

Ask three questions: Who is on the list? How long is my photo kept? Who reviews the alerts? If the answers are vague, the technology isn't watching criminals. It's watching you, and it will get you wrong eventually.

What would it take for a city to just... not do this? £320,000 buys a lot of street lighting. Share this with someone who thinks surveillance cameras are just "for safety."