Facial recognition technology keeps making the same mistakes

By Timilehin Ojo, INCLO Surveillance and Digital Rights Programme Manager

Since INCLO released Eyes on the Watchers, an in-depth report on the risks associated with the use of facial recognition technology (FRT) and guidelines to mitigate them, there has been continuous proof of the need for states and lawmakers to step up their game and review their use of the ubiquitous policing tech.

The trend observed since the launch of our report in February 2025 is simple: facial recognition technology keeps making the same mistakes. The question is no longer whether it fails. It is who bears the cost. Here are some of the latest examples from the United States, Ireland and Russia.

USA

The American Civil Liberties Union (ACLU) has continued to compile a publicly available report on these cases since 2019 to present. The last one on the list is Kimberlee Williams, the fourteenth documented wrongful arrest in the United States attributable to police reliance on erroneous facial recognition results. Kimberlee Williams had never been to Maryland. She lives in Oklahoma, where she was celebrating Christmas with her family when, hundreds of miles away, someone was committing bank fraud in her name, or rather, in her face. A bank investigator had run security footage through a facial recognition system and sent back Williams’ name as a match. Maryland police obtained arrest warrants. Nobody checked whether she could plausibly have been there. When she was finally arrested at a military base checkpoint during a routine DoorDash delivery with her daughter, she spent six months in jail before the charges were dropped. She was left on the street in December, with no phone and no money, halfway across the country from home.

UK-Ireland

Moving over away from the United States, a different kind of failure was playing out. In February, UK Border Force ran a live facial recognition pilot on the Dublin–Holyhead ferry route, scanning the faces of thousands of passengers against a watchlist of 6,535 suspected immigration offenders. The trial produced no matches. Across all exercises on the route since November, more than 10,000 faces have been scanned at a cost of at least £50,000 per deployment. Two people were arrested during the operations; the Home Office declined to say whether the facial recognition system had anything to do with it.

The findings raise fresh questions about the effectiveness and proportionality of the technology as both the UK and Ireland move toward wider deployment.

As ICCL’s Senior Policy Officer Olga Cronin told reporters in this related article, “How can it be necessary and proportionate to subject more than 10,000 individuals with no connection to wrongdoing to indiscriminate biometric facial data processing?”.

Russia

When we look at Russia, the story is not any different. On April 14, police at Moskovsky Station in St Petersburg detained Dmitry Galyanskikh, the frontman of a Russian rock band, after a facial recognition system flagged him as Ukrainian journalist Dmytro Gordon, who was sentenced in absentia to 14 years in prison in Russia for inciting terrorism. One officer checked his stubble and hair for signs of a disguise. Galyanskikh was eventually released.

Engines of injustice

A common trend from these INCLO jurisdictions is that facial recognition, when used as an investigative shortcut, operates as an engine of injustice. The technology’s higher error rates for people of colour compound pre-existing patterns of over-policing. The Dublin–Holyhead further affirms the position that mass biometric surveillance of people with no connection to wrongdoing cannot be justified under the guise of operational efficiency. The Russian case shows that the underlying logic of facial recognition, matching faces to databases, is always only as safe as the database itself, and that in authoritarian hands, the database fills with dissidents.

In the United Kingdom, Liberty Investigates has documented how these risks (explored in more detail in the report) continue to converge. Children as young as 12 appear on live facial recognition watchlists. Police deployed an algorithm with known demographic bias for years and lobbied to continue doing so. The UK’s first permanent facial recognition cameras went live in Croydon in October 2025. None of this has any primary legislative authority. The government’s public consultation on a legal framework for the technology ran until February 2026. The cameras, the vans, and the watchlists did not wait for it. The Metropolitan Police have used the technology 231 times to scan around 4 million faces in the last year. Despite the technology’s life-altering effects, the UK court delivered a disappointing decision in April 2026 in a challenge to the MET police’s use of FRT.

The fourteen cases on the ACLU’s list show a detailed empirical record of how consistently the technology interferes with individual fundamental rights in practice. Automation bias is not just a concept anymore; it can be seen in action in some of these cases. When the algorithm presents a face, officers go looking for confirmation rather than falsification.

Clear, rights-protecting and rights-preserving guidelines are still lacking in most jurisdictions where FRT is used. Structural accountability, voluntary disclosure, mandatory corroboration requirements with legal teeth, publicly available demographic audit obligations, a prohibition on watchlists composed of targets, and the suppression of evidence obtained through undisclosed FRT use are among the measures that must be implemented. This is a rule-of-law problem.

Sources

ACLU, “More Than a Dozen Wrongful Arrests Due to Police Reliance on Facial Recognition Technology,” See

The Detail, “Dublin-Holyhead facial recognition trial scans thousands but finds no matches,” See

Liberty Investigates, ongoing reporting December 2025, See; See

Meduza (English edition), “St. Petersburg facial recognition system mistakes rock musician for Ukrainian journalist,” See