Police watchdog warns facial recognition risks wrongful arrests
A UK inspectorate finds officers sometimes treat facial-recognition matches as proof rather than leads, raising the odds that innocent people — disproportionately from ethnic minorities — get wrongly arrested.
What happened: His Majesty's Inspectorate of Constabulary published a report on retrospective facial recognition (RFR) — the version police use after the fact to match images against databases, not the real-time kind used on the street. It found the tool is still error-prone and biased, more likely to wrongly flag people from ethnic minorities. Worse, some investigators didn't realise a computer match was only a lead, not proof, and acted on it without doing the follow-up checks required before an arrest.
Why it matters: When officers skip verification, the consequence isn't abstract: people can be arrested for crimes in places they've never been. Auditors also found poor-quality probe images — blurred photos, images with multiple faces — being fed into the system, further degrading accuracy. On top of that, custody photos that should have been deleted were still being kept and searched, meaning people never charged with a crime remain exposed to future false matches.
How it works, plainly: Retrospective facial recognition compares a photo from a crime scene or evidence against stored police images, mostly custody photographs, and returns a probability score for possible matches. It is meant to function as intelligence: a starting point that police must corroborate with other evidence before taking action. The report found that step was sometimes skipped, and that no single law currently spells out exactly how forces must use the technology or safeguard the public.
What needs to change: Inspector Kathryn Stone called for a clearer statutory framework, stronger national standards and better oversight to protect public confidence. Big Brother Watch's Jasleen Chaggar said officers are wrongly treating a probability score as evidence and that use isn't even being logged. The National Police Chiefs' Council defended the technology's crime-fighting value, while academic Pete Fussey noted that meaningful human review of AI matches is a legal requirement the report shows is often not being met.