'Dystopian' London Underground Face Scanners To Expand Despite 530,000 Failed Scans

British Transport Police is expanding live facial recognition across the London Underground after its trial scanned more than 530,000 faces without identifying a single wanted person.

Between February and July, BTP deployed the technology 18 times at major London railway stations. According to the force’s deployment records, those operations produced zero confirmed matches, zero arrests, and one false identification.

Despite those results, BTP expanded the trial to Tube stations on August 11, beginning at Victoria Underground station.^[1] The cameras will rotate between Underground and Network Rail stations, while the broader trial, initially scheduled to last six months, has been extended through November.

Silkie Carlo, director of privacy group Big Brother Watch, called it a “disturbing and dystopian expansion” that “treats the public like suspects.”^[2]

Over half a million faces were scanned

Live facial recognition cameras capture people’s faces as they pass through a designated area. Software then converts those images into biometric templates and compares them against a police watchlist.

BTP says those watchlists contain people wanted by the police or courts, including suspects linked to serious offenses and people accused of breaching bail conditions or court orders.

If the software detects a possible match, it generates an alert. A police officer then compares the camera image with the person in front of them and decides whether to approach the individual.

But during BTP’s first 18 deployments, the technology failed to generate one confirmed match.

BTP’s deployment register shows that individual operations ran for several hours at stations including London Bridge, Waterloo, Euston, King’s Cross, St. Pancras, Liverpool Street, and Victoria.

Each deployment compared thousands of commuters against watchlists generally containing several hundred people. Across the swathe of facial scans, the system generated only one alert, at King’s Cross in February.

It was false.

Additionally, BTP has not said whether any wanted people walked through the scanning zones during those deployments.

The failure is expanding anyway

BTP describes the move into the Underground as an extension of its trial and another opportunity to evaluate the technology.

“Expanding deployments into London Underground stations will help us assess the technology in a different transport environment while continuing to refine how it is used across the railway network,” Chief Superintendent Chris Casey said in an announcement.

Transport for London (TfL) says stations will be chosen using ‘crime data analysis.’ Its research found that 77% of the 150 girls aged 13 to 16 surveyed had experienced unwanted sexual behavior during a public transportation journey in the previous year.

“Live facial recognition has the potential to be a powerful additional tool in helping police quickly identify those wanted for high-harm offences,” Siwan Hayward, TfL’s director of security, policing, and enforcement, said.

What happens to your face after the cameras scan it

BTP uses NEC’s NeoFace M40 facial-recognition software. When someone enters a recognition zone, cameras capture their face and compare it against an authorized watchlist in real time.

BTP says biometric images that do not generate an alert are deleted automatically. Images connected to an alert must be deleted immediately after use or within 24 hours.

However, BTP also records the regular CCTV footage fed into the facial-recognition system and retains that footage for up to 31 days.

That creates several different categories of data, with different retention periods:

• Biometric images that do not match a watchlist are deleted immediately.
• Biometric images that generate an alert may be retained for up to 24 hours.
• Recorded CCTV footage from the facial-recognition cameras may be retained for up to 31 days.
• Information connected to an arrest or another police action may be retained under separate policing rules.

Immediate deletion reduces the risk of BTP creating a permanent biometric database containing every commuter who passes the cameras. However, members of the public still have to trust that the system deletes their biometric templates as promised, uses appropriate watchlists, and does not expand its purpose later.

Facial geometry is persistent biometric information that can potentially be used to identify a person across different images, cameras, and databases.

That concern is not limited to the U.K. In the U.S., ICE is seeking technology capable of identifying people through facial images, device data, behavioral signals, commercial records, and publicly available photographs.

BTP acknowledges that some faces can be harder to match

BTP says its facial-recognition algorithm was independently tested and selected partly because of its performance across different demographic groups.

The force operates the system at a minimum threshold setting of 0.64, above the 0.60 setting at which its testing reportedly found no statistically significant differences among the tested groups.

However, BTP’s own equality impact assessment acknowledges circumstances that can reduce accuracy.

The document says older custody images may no longer reflect a person’s current appearance. It also notes that facial-recognition accuracy can be lower for children, particularly those under 13, because their faces change as they grow.

BTP further acknowledges possible effects on transgender, nonbinary, and gender-fluid people whose current gender presentation differs from their reference images.

These risks matter because a false alert can cause an innocent person to be approached and treated as a possible criminal in public. Human review may sound reassuring, yet it can invite cognitive offloading and automation bias: The software has already done the initial “thinking,” and the officer begins with a machine-generated suspicion that may shape what they see next.

Once the system labels someone a possible match, the reviewer may look for reasons to confirm the alert instead of evaluating the person neutrally. Misidentification can influence the encounter before a word is ever exchanged.

Can commuters avoid the facial-recognition cameras?

BTP says deployments will be clearly marked, and people who do not want to enter a recognition zone will have an alternative route available.

Travelers should look for signs identifying where live facial recognition is operating and ask station staff or police for the alternative route before entering the marked area.

There is little that ordinary digital privacy software can do against a physical facial-recognition camera. That leaves commuters dependent on visible warnings, accessible alternative routes, deletion rules, human review, and police compliance with internal safeguards.

Facial recognition is becoming part of everyday public life

The Tube trial is part of a much broader expansion of facial recognition in public and commercial spaces.

Police forces are increasingly using the technology in transportation hubs, shopping districts, and large public events. Private businesses are also experimenting with facial scans to identify suspected shoplifters, control access, enforce purchase limits, and verify customers’ identities.

The experiment is now moving deeper into London’s transportation network without a single successful identification to demonstrate its effectiveness.