What is AI facial recognition in policing?
Facial recognition technology works by mapping the geometry of a person's face — the distance between the eyes, the width of the nose, the shape of the jawline — into a numerical fingerprint that can be compared against a database of stored images at speed. In a policing context, that database is typically built from custody photographs, driving licence photos, or purpose-built watchlists compiled for specific operations. When the algorithm finds a potential match, it flags the result for a human officer to review. The technology does not make arrests on its own; it generates a lead.
Two quite different deployment models dominate the conversation. Live facial recognition — sometimes called LFR — scans the faces of people moving through a public space in real time, usually from a camera mounted on a van or a fixed position on a building, and alerts officers the moment someone who matches a watchlist entry walks past. Retrospective facial recognition, by contrast, takes a single still image — often grabbed from CCTV footage after a crime has already taken place — and runs it against a database to try to identify a suspect. The ethical and legal questions raised by each are rather different, and it is worth keeping that distinction in mind when reading coverage, because the two are frequently conflated.
Where is it being used in the UK?
The Metropolitan Police has been the most prominent and arguably the most controversial adopter of live facial recognition in the United Kingdom. The Met began trialling the technology in public in 2016 and has progressively expanded its operational use ever since, deploying cameras at events, high streets, and transport hubs across London. South Wales Police ran its own extensive trial programme, which resulted in a landmark legal challenge in 2020 when the Court of Appeal ruled that the force's use of the technology was unlawful — not because facial recognition itself was necessarily off-limits, but because the legal framework governing its use was insufficiently precise and the safeguards around who could appear on a watchlist were inadequate.
Other UK forces have been somewhat more cautious, though retrospective facial recognition has become a quietly routine investigative tool in forces that would not describe themselves as early adopters of the technology. The National Police Chiefs' Council has published guidance, but there is no single piece of primary legislation in England and Wales that specifically regulates police use of facial recognition — a gap that campaigners have repeatedly highlighted and that the government has so far declined to fill.
Scotland operates under a different legal framework, and Police Scotland has taken a noticeably more restrained approach to live deployment than forces south of the border, though the use of retrospective tools is similarly widespread and largely uncontroversial within the service itself.
Accuracy — and who it gets wrong
The question of accuracy in facial recognition is rather more complicated than vendors tend to let on. The headline figures cited in sales materials — often 99-plus per cent accuracy — are typically measured under controlled conditions, with high-quality frontal images, good lighting, and a relatively small gallery of faces to search. Real-world policing conditions are rarely that obliging. CCTV footage is frequently low resolution and poorly lit, subjects are moving and not looking at the camera, and the databases being searched can run into millions of records.
The more significant concern, however, is not overall accuracy but differential accuracy. Large-scale independent testing — most notably by the US National Institute of Standards and Technology, which has assessed more than a hundred algorithms — has found that many facial recognition systems perform materially less accurately on women, on people with darker skin tones, and on older adults than they do on white men in their thirties and forties. The gap between the best and worst performers on demographic subgroups is not marginal; in some cases it runs to an order of magnitude. This disparity matters enormously in a policing context because a false match does not merely mean a missed identification — it can mean an innocent person being stopped, questioned, or in the worst cases arrested.
There have been documented cases in the United States of wrongful arrests directly attributable to flawed facial recognition matches, and civil liberties groups in the UK have catalogued a series of cases where the Met's live system generated alerts that led to individuals being stopped who turned out to have no connection to the watchlist subject. The Met has maintained that its human review process filters out the vast majority of erroneous matches before any stop occurs, but independent observers have pointed out that the very act of being stopped on a busy high street is itself an intrusion, regardless of whether anything follows from it.
The legal landscape
There is no single, tidy answer to whether police facial recognition is legal in the UK — it depends on which force, which deployment model, and which data protection framework you are asking about. What is clear is that the law has not kept pace with the technology.
The 2020 Court of Appeal ruling in R (Bridges) v Chief Constable of South Wales Police is the most significant judicial intervention to date. The court found that South Wales Police's use of automated facial recognition breached the Human Rights Act, data protection law, and the Public Sector Equality Duty — not because LFR is inherently impermissible, but because the specific legal framework within which the force was operating lacked sufficient clarity about who could be put on a watchlist, what criteria governed its use, and how the equality implications had been assessed. The ruling has shaped how other forces approach their legal justifications, but it has not resolved the fundamental question of what a lawful statutory framework for LFR would look like.
The Information Commissioner's Office has raised concerns about the lawfulness of various deployments and has issued guidance stressing that biometric data — which facial recognition images constitute — carries the highest level of data protection obligation. Meanwhile, Parliament has debated the issue on several occasions without legislating, and the current government's position is broadly supportive of police use of the technology provided appropriate safeguards are in place, though what those safeguards should be remains contested.
In the United States, a patchwork of local bans and restrictions exists in cities including San Francisco, Boston, and Portland, but there is no federal regulation, and use by federal law enforcement agencies including the FBI and Customs and Border Protection is widespread. The European Union's AI Act, which came into force in 2024, imposes significant restrictions on real-time biometric identification in public spaces, though with exceptions that critics have argued are broad enough to largely preserve existing police capabilities.
The watchlist problem
One aspect of facial recognition that receives less attention than the technology itself is the question of who ends up on the watchlist being searched. In a live deployment, the camera is scanning the faces of everyone who walks past — the vast majority of whom are entirely innocent and have no idea they are being checked. Whether that matters constitutionally and ethically turns partly on how the watchlist was compiled.
If a watchlist is tightly defined — wanted individuals with active arrest warrants, for instance — the privacy intrusion on the innocent majority is arguably modest. But watchlists compiled for major public events have in some cases included individuals who are not suspected of any crime but who are considered subjects of intelligence interest, protest organisers, or people subject to civil court orders. The Bridges judgment specifically criticised the absence of clear rules about who could appear on a South Wales Police watchlist, and this remains an unresolved tension in current UK deployments.
Follow the coverage
PoliceAI News tracks every facial recognition story from across the English-speaking world as it breaks — new deployments, court rulings, parliamentary debates, academic research, and community responses. The feed refreshes every 30 minutes.
View Live Facial Recognition StoriesYou can also browse every facial recognition story in the full archive, or explore related topics: predictive policing algorithms and AI-equipped body-worn cameras.