Society & Ethics

Ethnic minorities four times more likely to face police AI face scans in Wales

South Wales police data shows ethnic minorities, 7% of the local population, received 24% of AI face scans in 2025, prompting bias warnings from campaigners and the force itself.

Ethnic minority groups in Wales four times more likely to be target of police facial scans
Ethnic minority groups in Wales four times more likely to be target of police facial scansgeralt / pixabay
By Marcus Bennett6 min read

Why it matters

  • Ethnic minorities made up 24% of police face scans in south Wales in 2025 despite being 7% of the population.
  • South Wales police scanned 257 adults and 14 children, including two under-10s, in 2025.
  • Arrests followed in 48% of cases where the Neoface algorithm produced a match.
  • Alvi Choudhury, 26, was wrongfully arrested in February after the software confused him with another man.
  • A Croydon live facial recognition pilot led to 173 arrests between October 2025 and March 2026.

Ethnic minority people in south Wales were selected for police facial recognition checks at four times the rate of their white counterparts during 2025, according to internal police data obtained under the Freedom of Information Act and shared with the Guardian. The figures show that 24% of people chosen by South Wales police officers for face scans were from ethnic minorities, a group that makes up only 7% of the area's total population. Civil liberties campaigners called the numbers "alarming," and the force itself conceded they raise "concerns about disproportionality."

The disproportionality persists even when measured against arrest statistics for the same police area. People from ethnic minorities account for 11% of arrests by South Wales police. The face-scan rate of 24% stands at more than double that benchmark, and more than triple the group's share of the population. The data covers a full year of use in 2025.

South Wales police polices Cardiff and Swansea, Wales's two largest cities, and has pioneered several versions of AI-enabled facial recognition. The latest figures concern "officer-initiated facial recognition" (OIFR): a police officer captures a person's face with a smartphone and runs the image through an algorithm called "Neoface" to match it against custody mugshots and other face databases. Most scans occur because the person is suspected of a crime.

Why the numbers matter for AI policing

The stakes extend beyond one constabulary. South Wales is one of the first UK police forces to routinely deploy OIFR, and the technology is spreading. The Metropolitan police in London is trialling a similar system, and it has also been tested in Liverpool. Critics have labelled facial recognition tools "Orwellian" and have repeatedly warned that algorithmic bias puts ethnic minorities at greater risk of wrongful arrest. Last week, the police watchdog warned that innocent people, especially from ethnic minorities, remain at risk of wrongful arrest if police treat inaccurate retrospective facial recognition data as evidence.

Advocates, including Home Secretary Shabana Mahmood, counter that facial recognition is the biggest breakthrough for catching criminals since fingerprints and DNA matching. That endorsement from the top of government, combined with expanding deployments, makes the South Wales disproportionality data a key test case for whether the technology can be used fairly at scale.

What does the 2025 data show in detail?

The full-year breakdown of OIFR use by South Wales police includes:

  • 257 adults and 14 children were scanned, including two children under the age of 10.
  • More than half of scans involved people suspected of a crime, being unlawfully at large, or breaching bail conditions.
  • The next most common reason was officer belief that the person posed a risk of harm to themselves or others, such as stalkers or sex offenders.
  • About one in ten scans targeted a suspected missing person.
  • The technology was used to identify a dead person roughly once a fortnight.

The technology also serves a stated welfare purpose: identifying suspects, wanted criminals and missing people who decline to give their details when challenged, as well as people found unconscious, uncommunicative or dead.

Outcomes from the scans cut both ways. In many cases the scan failed to deliver a match, or the subject agreed to share their identity. Where a match was made, arrests followed in 48% of cases. No further action resulted in a third of cases.

Who is speaking out, and what did they say?

The figures were obtained by Big Brother Watch, a civil liberties campaign group. Jake Hurfurt, the organisation's head of research and investigations, said: "This data shows that intrusive biometric tools are being used to disproportionately target minority communities."

"It is alarming that non-white people have been subject to face scans at a much higher rate than would be expected either by the region's demography or the makeup of people arrested in the region," Hurfurt said.

Aliya Mohammed, chief executive of Race Equality First, a Cardiff-based anti-discrimination organisation, said the figures raised "serious concerns about racial disproportionality."

"Facial recognition is not neutral because it uses AI," Mohammed said. "If existing biases influence who is stopped and scanned, technology can reinforce and even amplify inequality."

The data also resonates with people directly affected by the technology's failure modes. Alvi Choudhury, 26, was arrested in February for a burglary committed in a city he had never visited, after police face-scanning software confused him with another person of south Asian heritage on a custody database. He said it was "very concerning" that ethnic minorities were being disproportionately scanned.

"The police are under pressure to solve crimes and they are over-reliant on these AI tools," Choudhury said. "They shouldn't be arresting people based on the AI tools alone."

Choudhury said he remained worried that AI systems were not trained on enough ethnic minority faces to ensure the rate of false positives was no greater than it is for white people. His case is a concrete instance of the wrongful-arrest risk the police watchdog flagged last week.

How did South Wales police respond?

A spokesperson for South Wales police said: "We recognise the figures raise concerns about disproportionality. The Number of OIFR searches remains relatively small, and the use of the technology is in its infancy. That said, each deployment is based on the individual circumstances that officers are dealing with at that time and is subject to supervisory and force-wide scrutiny."

The force said it would "continue to review our approach through our established governance processes, which also include input from independent advisory groups and external scrutiny panels." It added: "Where opportunities for learning or improvement are identified, we will take appropriate action."

What oversight is coming next?

Regulatory pressure is building. Earlier this month, the Home Office announced a new facial recognition watchdog with the power to stop police forces from using the technology if they fail to follow rules and new laws set out in the police reform bill. Those laws include a duty for the use of face scans to be intelligence-led and proportionate — a standard against which the South Wales disproportionality figures will now be measured.

The Home Office said decisions on the operational use of facial recognition were matters for the police, but that forces must comply with equality, human rights and data protection laws, and that the technology's use must be necessary, proportionate and fair.

Other versions of the technology are already producing results that proponents cite in its defence. Live facial recognition cameras, directed from marked vans and from cameras fixed to lamp-posts, were piloted in Croydon between October 2025 and March 2026. That pilot led to 173 arrests, including of rape, kidnap and sexual assault suspects.

The South Wales data lands at the intersection of those two trajectories: expanding deployment and tightening oversight. With the Metropolitan police trialling the same officer-initiated system, the new watchdog's first tests may well turn on whether forces can close a gap in which 24% of scans fell on 7% of the population.

Source: theguardian.com

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

613 articles

Related articles

  1. WA police scanned 1.38 million faces, made 122 arrests and logged nine false alerts
  2. AI Embedded Across UK Home Office in Near-Total Secrecy, Report Claims
  3. Adversarial Fashion Takes Aim at AI Surveillance Cameras
  4. A Selfie to Predict Cancer Survival: 40 Million Faces, and Reason for Doubt
  5. Flashing Lenses in the Senedd: Smart Glasses Reignite Ban Calls

« Previous article