Britain’s police watchdog has drawn a line that every force using facial recognition — including in New Zealand — will be asked to draw in their own countries: a match generated by the software is a starting point for an investigation, not evidence that the person on the other end of the door committed a crime.
The warning comes from His Majesty’s Inspectorate of Constabulary and Fire & Rescue Services (HMICFRS), whose report on how police forces and the National Crime Agency use retrospective facial recognition (RFR) was published this week, as reported by The Guardian on Thursday, 12 November. RFR, the less celebrated sibling of live facial recognition, compares images captured during investigations — CCTV stills, seized phones, crime-scene frames — against millions of photographs held on the Police National Database. Inspectors examined seven forces plus the NCA, in fieldwork carried out in March and April, according to the watchdog’s own reporting as carried by the Dorset Echo on Yahoo News UK.
What they found cuts to the heart of how AI is quietly rewiring police work.
The computer said it was him
The central finding is deceptively simple: some investigators did not know that RFR results are intelligence rather than evidence, and therefore “required further enquiries before officers took enforcement action.” In a small number of cases the report goes further, saying officers “relied too heavily on RFR results, which led to inappropriate investigative action.”
That is inspectorate language. Translated: police arrested or moved against people primarily because an algorithm matched their face — and without enough human verification before the knock came to the door.
The report also found the technology carries precisely the bias problem critics have documented for years, saying more needs to be done to reduce the risk of people being wrongly arrested, with the algorithm more likely to wrongly flag people from ethnic minorities as suspects. Auditors found poor-quality probe images — blurred frames, photos containing multiple faces — being fed into search systems, degrading match reliability. And custody images that should have been deleted under retention rules remained available for facial recognition searches, meaning people whose photographs had no lawful place in the system could still be matched against new crimes.
The structural problem the inspectors describe will sound familiar to anyone who has watched a government deploy cutting-edge AI on century-old legal scaffolding: “legislation hasn’t kept up with changes in facial recognition technology and there is no comprehensive statutory framework governing this technology’s use.” Forces, in other words, are guessing at their own lawful basis.
The audit numbers say the oversight is mostly notional
The most quietly damning passage concerns the police’s own audit regime. According to the Mirror’s syndicated reporting, NPCC figures show only two forces in England and Wales had audited the required number of facial recognition records. Seven forces had not appointed an auditor at all, and 22 had failed to submit an audit return or filed one that was incomplete.
That is not a system under control. That is a system that has been scaled faster than its supervision — police forces carry out more than 25,000 facial image searches every month on the Police National Database, per the Home Office’s own factsheet, and the statute to govern all of it still does not exist.
The police response is split into two familiar registers. Jasleen Chaggar of Big Brother Watch called it alarming that officers treat facial recognition as bona fide evidence rather than a probability score, noting people could become suspects “without even realising why they’ve been accused.” The National Police Chiefs’ Council’s facial recognition lead, Ben Russell, welcomed the report’s recognition of the technology’s “exciting possibilities,” per The Guardian, saying policing remains committed to “ethical, proportionate and responsible use.”
Professor Pete Fussey of the University of Southampton, who has spent years inside police forces as an independent AI observer, put the sharpest version on record: meaningful human scrutiny of AI systems is a legal requirement — and the report documents investigators ignoring it.
Why this lands in New Zealand
New Zealand has its own history here — NZ Police trialled facial recognition technology without clearance, and the warning that a Māori person would one day be wrongfully arrested on a false match has been on the record from researchers for years. But the UK report names the exact failure mode that makes policy assurances worth little without structure behind them. It is not that officers are careless. It is that the default gravity of an AI tool — a confident match, in seconds, against millions of faces — pulls the human in the loop toward rubber-stamping. The UK’s inspectors found the gravity winning even in trained investigators, in forces that had signed on to safeguarding frameworks.
There is also a second lesson in the report, one about deletion. Custody images that “should have been deleted” were still in the search pool. In a country where police image retention has already been flagged by oversight bodies, the UK experience shows retention rules with no enforcement become search databases by drift — not by decision, policy or law.
The UK is ahead of New Zealand only in the sense that it got there first. The Inspectorate’s real finding is that deploying RFR without a statutory framework, enforced audits and mandatory corroboration produces a predictable pattern of harm concentrated on ethnic minorities — and by the time a watchdog writes that pattern down, the databases are already built, and the people flagged already know.
Sources: The Guardian — Facial recognition errors put ethnic minorities at risk of wrongful arrest, police watchdog warns, Dorset Echo syndicated on Yahoo News UK — Watchdog warns facial recognition errors could lead to mistaken arrests, GOV.UK — Police use of facial recognition factsheet, ICO — Why data protection lies at the heart of responsible police use of facial recognition
