· 10 min read · by PoliceAI News

Facial recognition has moved from the van to the officer's phone

The Metropolitan Police began a six-month pilot on 17 November that lets 100 officers photograph a person during a stop and check the image against about a million custody photographs. It is the third kind of police facial recognition, and the one with the least written law around it. The Met's own policy is detailed. What the pilot needs now is a public record of how it is used.

For most of the past decade, the public picture of police facial recognition has been a van with cameras on its roof and signs on the pavement telling passers-by that their faces are being compared with a watchlist. That picture now needs a third element. On 17 November the Metropolitan Police began a six-month pilot in which 100 officers can use an app on their phones to photograph someone they have stopped and compare the image with a database of around one million custody photographs, a capability known in policing as operator-initiated facial recognition, or OIFR.

The coverage this week, led by the fact that officers do not need a person's consent to take the photograph, was accurate on that point. It is also worth setting out what the tool is meant to do, what the Met's own rules for it say, and what the forces that adopted it earlier have learned, because the pilot is likely to shape how the rest of England and Wales uses it.

What the tool is for

The Home Office describes three kinds of police facial recognition. Retrospective searches compare an image from CCTV, a doorbell or a phone with custody photographs after an event. Live facial recognition compares the faces of everyone passing a camera with a watchlist in real time. Operator-initiated facial recognition sits between the two: in the Home Office's words, it is a mobile app that allows officers, after engaging with a person of interest, to photograph them and check their identity where they are not sure, without having to arrest them and take them into custody.

That last clause is the operational case for it, and it is a real one. An officer who stops someone suspected of an offence and is given a name that cannot be checked currently has a limited set of options, and one of them is to arrest the person so that their identity can be established at a custody suite. That takes the officer off the street for hours and the person stopped into a cell. Commander Nick John told the BBC that if the app means "a far speedier response than we would ordinarily" get, that is a good thing for the public of London, and that any technology which helps officers understand who they are dealing with is worth testing. He also said that officers would not be going around London taking photographs from a distance of people who are not suspected of involvement in any offending.

The Met's policy, which it has published with a legal mandate, a data protection impact assessment and an equality impact assessment, also allows the app to be used to assess whether someone who has been arrested or detained poses a risk to themselves or to others. In practice that covers people who are unable or unwilling to say who they are and who may be vulnerable, which is a use few would object to.

What the Met's rules say

The policy is more detailed than most of the reporting suggested, and several of its safeguards go further than the public has been told:

  • The only database searched during the pilot is the Met's custody image collection. A separate store of images from unsolved crimes will not be switched on during the proof of concept.
  • A photograph that produces no further lines of inquiry is deleted from the system within 31 days of the search.
  • Possible matches are shown as a ranked list. The policy says officers must not assume that a highly ranked result is a correct one, and that any match is an intelligence lead rather than confirmation of identity for evidential purposes.
  • No authority is given to photograph children believed to be under 13, and searches on anyone under 18 need the approval of an inspector. The Met has also said that under-18s will be excluded entirely for the first four weeks.
  • The app must not be used in a way that amounts to directed or intrusive surveillance, and the policy asks officers to consider the effect on freedom of expression and assembly where a photograph features people at a lawful protest.
  • Every search, including those that produce no viable match, needs a written match report, and the system keeps an audit log that cannot be altered once written.
  • The Met's facial recognition technology board will review use at least every quarter, the Mayor's Office for Policing and Crime, the London Policing Ethics Panel and the Information Commissioner's Office provide external scrutiny, and the force has committed to an annual report on its use of OIFR.

The point that drew the headlines is in the policy too. The photograph is taken either under statutory powers, where someone has been arrested or detained, or otherwise with the person's consent or under the common law power to take a photograph from a lawful vantage point without compulsion. Consent should be sought before that common law power is used, but, as the policy puts it, if consent is withheld this "does not of itself prevent the taking of a photograph". Officers must tell people that a photograph is being taken, and the Met has said they will not use force if someone covers their face.

What Wales has already seen

The Met is not the first force to do this. South Wales Police and Gwent Police published a joint OIFR policy and legal mandate for a pilot in 2024, after an earlier phone trial in 2021 and 2022, and their officers use the capability through an app called iPatrol. The Information Commissioner's Office looked at the pilot during its audit of the two forces in 2025 and gave them feedback on it. Liberty Investigates and The Times reported in May that Merseyside Police had also given officers access in a pilot, that West Yorkshire Police expected to deploy it in 2026, and that Thames Valley Police had explored adopting it, with documents indicating that the capability would be offered through Motorola's Pronto policing app.

The Welsh figures, which Big Brother Watch obtained through a freedom of information request and which were reported last week, are the best evidence so far of how the tool behaves once it is in routine use. South Wales Police used it 271 times in 2025, on 257 adults and 14 children, two of them under ten. Where a match was made, an arrest followed in 48 per cent of cases. People from ethnic minority backgrounds made up 24 per cent of those photographed, against 7 per cent of the population the force serves and 11 per cent of the people it arrests.

Those numbers can be read more than one way, and it is fair to say so. The comparison with arrests narrows the gap considerably, 271 uses is a small sample, and the force's response was measured: it said the figures "raise concerns about disproportionality", that use of the technology "remains relatively small" and "is in its infancy", and that it would review its approach with independent advisory groups and external scrutiny panels. That is the kind of answer that only becomes possible when the figures exist in the first place, and it is the strongest argument for the Met's pilot producing the same figures from the start.

The concerns, stated fairly

Civil liberties groups have objected to the pilot on principle as well as on its detail. Akiko Hart of Liberty said that OIFR "combines all of the problems of stop-and-search with those of facial recognition". Jasleen Chaggar of Big Brother Watch argued that people generally do not have to give their identity to the police unless they have broken the law, and that the app gives officers a way of identifying people that they have not had before. Habib Kadiri of StopWatch said there was no evidence that enough had been done to mitigate bias.

Those arguments deserve to be weighed against the Met's, rather than set aside. The policy limits use to people an officer reasonably suspects of a recordable offence, people who are wanted, and people who have been detained, which is narrower than the "stop and scan" description implies. Against that, a recordable offence covers a great deal of ordinary policing, and the fairness of the outcome depends heavily on who officers choose to stop, which is exactly the question the Welsh figures raise.

There is also a question of openness that the Met itself would probably accept. In February the force's website still said it did not use operator-initiated facial recognition, and the pilot became public only when the Mayor confirmed it in answer to questions from the London Assembly member Zoë Garbett. The Met has since published a full set of governance documents, which is a substantial improvement, but the episode explains some of the suspicion that greeted this week's launch.

Where the law has got to

On 4 November the Home Office published its plans for a statutory framework covering police use of facial recognition, with an independent regulator and a code of practice, to be brought in through the Police Reform Bill. We covered those plans when they were announced. As far as we can find, the government's response defines facial recognition as including "both live and retrospective uses" and does not mention operator-initiated recognition by name, although its broad scope would probably cover it. The Home Office told Liberty Investigates earlier this year that it had no plans to roll the capability out nationally, while the same reporting found national policing papers discussing how it might be offered across forces.

Until the Bill passes, the Met's pilot rests, like the Welsh one, on the common law, the Data Protection Act and the force's own policy. That does not make it unlawful, and the Court of Appeal's 2020 judgment in the Bridges case against South Wales Police turned on the gaps in a force's own rules rather than on the technology itself. It does mean that the policy documents are, for now, the rules, which makes how faithfully they are followed, and how openly that is reported, unusually important.

What the pilot should publish

The Met has promised an annual report, but the pilot ends in May and a decision on wider use will follow. A short set of figures, published at the end of the six months, would allow that decision to be judged on evidence rather than on assertion from either side:

  • the number of searches, broken down by the reason given and by whether the person had been arrested or detained;
  • how many searches produced a match that an officer accepted, and how many of those were later found to be wrong;
  • the outcome in each case, including arrests avoided because an identity was confirmed on the street;
  • the age and ethnicity of those photographed, compared with the people the Met stops and arrests;
  • how often consent was sought, how often it was refused, and how often a photograph was taken anyway.

The last of those figures matters most. A tool that is mainly used with people's agreement, to save them a trip to a custody suite, is a very different thing from one that is mainly used over their objections, and at the moment nobody outside the Met can say which this will be. The force has written careful rules. Publishing how they work in practice would be the most persuasive case it could make for keeping the tool.

The pilot is on the tracker as a Metropolitan Police trial, alongside the existing records for South Wales Police, Gwent Police, Merseyside Police and West Yorkshire Police. Our facial recognition guide explains the three kinds of police facial recognition in more detail.

All analysis

More analysis

All analysis
  • The facial recognition searches police cannot account for

    Police in England and Wales ran more than 290,000 retrospective facial recognition searches last year. The policing inspectorate has now reported that forces cannot show what those searches achieved, that most are not auditing them as required, and that some of the photographs being searched should have been deleted. The coverage led on the risk of wrongful arrest. The record-keeping is the larger

  • Britain's new facial recognition regulator stops at the border

    On 4 November the Home Office set out plans for a statutory framework and an independent watchdog with the power to stop a force using facial recognition. Two days later Police Scotland demonstrated the technology at its own headquarters. Policing is devolved, the new regulator consolidates two offices that do not exist in Scotland, and Scottish ministers face a statutory deadline on 30 November.

  • The backlash against licence plate cameras did not start with the cameras. It started with the search logs.

    In eight days, Florida revoked permits for licence plate readers on state highways, a bipartisan bill was introduced to strip federal funding from them, and a Senate investigation widened to three more suppliers. The cameras have been going up for years without much argument. What changed is that the record of who searched them, and why, became readable by the public.