London scans 500,000 faces with no direct arrest

What this is about

A British Transport Police live facial-recognition trial at London stations scanned more than 500,000 faces between February and July 2026. Documents obtained by Liberty Investigates through freedom-of-information rules and shared with the Guardian show that 18 deployments produced one alert against a watchlist. It was a false match, and there was no direct arrest.

The six-month trial cost £320,786 for equipment and police staffing and used almost 100 officer hours. It was nevertheless extended for four months in August and expanded to selected London Underground stations. The case provides unusually concrete figures for a debate that often remains abstract.

What live facial recognition actually does

Cameras capture the faces of passers-by and compare calculated features in real time with a prepared watchlist. If the system reports a match, a human reviews the alert and decides what happens next. This is more than ordinary video recording: biometric identifiers are a specially protected category of personal data.

Police say they adjusted watchlists, locations, and procedures during the trial. In the extended phase, three confirmed alerts were produced, but those people were complying with their court-imposed conditions. Other arrests made during deployments were not directly caused by a facial-recognition alert and therefore are not part of the system's performance record.

Why it matters

More than half a million people were biometrically checked even though almost nobody on the relevant list was identified. The balance between intrusion and outcome is therefore central. A system can be accurate in a laboratory yet produce little value in practice when watchlists, location, and timing do not align with the movement of wanted people.

Use is also expanding. According to the Guardian, more than half of police forces in England and Wales have deployed live facial recognition. London plans fixed cameras in the West End, and the redesigned Oxford Street is also due to be covered. Without clear legal boundaries, isolated trials can become permanent infrastructure before benefits, oversight, and complaint routes are settled.

In plain language

Imagine checking 500,000 train tickets because one wanted person might be among them. After every check, an uninvolved person is wrongly pulled aside and the wanted person is not found. Even if the scanner is technically accurate, the result raises a question about whether the effort and intrusion were proportionate.

A practical example

At a busy station, 30,000 people pass the cameras during one deployment. A watchlist contains 200 people. The system flags one traveler because her calculated facial profile resembles a listed image. An officer checks her identity and the image and resolves the mistake after two minutes.

The intrusion is real for the traveler even though she can continue. Police incur review and documentation work. For the other 29,999 people, the question remains whether processing their biometric data was necessary when no match resulted. Any expansion must justify that proportionality.

Scope and limits

• The published figures assess this specific trial, not every form of facial recognition in every setting.
• No direct arrest does not mean that no policing or unrelated arrests occurred during deployments.
• One false alert across more than 500,000 scans says little about a false-match rate if very few listed people actually passed the cameras.

The decisive factors are therefore not only accuracy figures but also watchlist selection, deployment location, independent oversight, deletion periods, and an effective route for people to challenge misuse.

SEO & GEO keywords

live facial recognition, London, British Transport Police, railway stations, biometric data, surveillance, false match, privacy, Liberty Investigates, London Underground, police technology