Justice

UK police still using custody images that should have been deleted, watchdog finds

Custody photographs that police should have deleted remain available for retrospective facial recognition searches, HMICFRS has found. Its inspection also identified gaps in oversight and officers' understanding of a technology that UK law enforcement uses for more than 300,000 searches a year.

lake-uankxtpl20t15euqi6743503-7095eb62-f417-4e56-82bb-60ec13af5486-0.webp

Retained images remain searchable

The police inspectorate's 58-page report, published in November 2026, examined retrospective facial recognition (RFR) use by police forces and the National Crime Agency (NCA). Home Secretary Shabana Mahmood commissioned the work a year earlier.

Inspectors found that police were keeping custody images in breach of rules governing their review, retention and disposal. Although deletion practices fell largely outside the inspection's focus on governance and operations, HMICFRS established that images due for deletion were still being searched. It warned that this could damage public confidence. Potentially millions of unlawfully retained photographs remain at issue.

RFR searches an existing collection of photographs for possible matches to an image investigators already hold. It differs from live facial recognition, which compares faces against a predetermined watchlist in real time. Police use RFR to compare an unidentified person's image with custody photographs taken after arrest.

The Home Office introduced facial searching on the Police National Database (PND) in 2013 for trained, licensed users seeking to identify unknown people of interest. The database contains more than 20 million facial images among billions of records assembled from several hundred databases. Contributors include all 43 forces in England and Wales, Police Scotland, the Police Service of Northern Ireland, the NCA and other law enforcement bodies.

The inspection's fieldwork ran from March 3 through April 30, 2026, covering the NCA and seven forces: British Transport Police, Cheshire Constabulary, Dorset Police, Gwent Police, the Metropolitan Police, Northumbria Police and Surrey Police. Inspectors reviewed 416 documents and conducted 91 interviews and focus groups.

Potential matches mistaken for evidence

HMICFRS found substantial differences in how forces conduct searches and oversee their use. Public consultation, transparency and record-keeping were inconsistent within a legal and regulatory system the inspectorate described as fragmented and underdeveloped. Forces also often lacked the management information needed to assess RFR's contribution to investigations or demonstrate improvements in performance.

Some forces specified how officers should search, record results and seek corroboration; others had no formal procedures. Investigators and supervisors often omitted RFR considerations and searches from investigation plans, inquiry logs and supervisory reviews. Those omissions left forces without reliable records of how a result influenced a decision or whether officers had checked it against other information.

Inspectors also found that some investigators did not understand that an RFR result is intelligence, not evidence, and requires further inquiries before enforcement action. Some failed to do enough corroborative work. In a small number of cases, excessive reliance on a potential match led to inappropriate investigative action. HMICFRS warned that such practices risked incorrect identifications or arrests.

The report's search statistics show that results commonly offer several possible candidates rather than one. PND data for the year ending March 31, 2026, showed that 74% of submitted RFR searches produced at least one face match. Searches returning matches usually generated multiple candidates, averaging between nine and 15 per search.

A separate Home Office survey, published November 12, 2026, found similar gaps in users' understanding. Among respondents, 34% either incorrectly believed that a PND facial search match met the evidential requirements for recognition under PACE Code D or were unsure whether it did. Another 65% could not remember receiving training during the previous year.

The survey reported that 6,012 distinct users conducted more than 300,000 PND facial searches between February 2025 and January 2026, or about 25,000 a month. Despite the knowledge and training gaps, 79% of respondents described their experience as good or very good.

Accuracy, bias and public confidence

Testing also exposed trade-offs in the system's settings. A 2024 National Physical Laboratory assessment of the PND's RFR algorithm found that, at some settings, the likelihood of incorrect inclusion in search results differed by demographic group. Women were more likely to be included than men, Asian or Black people more likely than White people, and Black women more likely than Black men. People ages 21-29 were also more likely than other age groups to appear on candidate lists.

After the operator changed the system's sensitivity, the demographic differences were no longer statistically significant. But the adjustments also produced fewer results and lowered the likelihood of finding a true match.

HMICFRS inspector Kathryn Stone said public support depended on police demonstrating responsible use. "Police forces must be able to show that appropriate safeguards are in place, risks relating to bias and data retention are effectively managed, and decisions are still subject to human judgment," she said. "A clearer statutory framework, stronger national standards and improved governance and oversight arrangements are essential to maintaining public confidence."

The report also cited the Home Office survey "Public attitudes to police use of facial recognition technology," published in December 2025. It found that 64% of respondents supported police use of facial recognition. Of the three capabilities examined, RFR recorded the highest public acceptability, at 97%; acceptability for live facial recognition ranged from 88% to 91%, depending on its purpose.

An unresolved retention problem

The custody-image issue predates the inspection by more than a decade. In a 2012 ruling not mentioned in the report, the High Court found retention of custody images in the PND unlawful because records concerning people who had not been convicted were treated like those concerning convicted people. It also found the six-year retention period disproportionate. Successive biometric commissioners have since warned that millions of images remain unlawfully held.

In February 2022, Lee Freeman, the National Police Chiefs Council (NPCC) lead for records management, raised the issue with fellow chief constables. He warned that potentially unlawful retention of millions of photographs threatened police legitimacy, exposed forces to litigation and risked challenges to their use of the images for facial recognition.

HMICFRS said it was reassured by Programme Tabula, launched by the Home Office and NPCC in October 2023 to address custody-image retention. The inspectorate welcomed the project and said it wanted to learn how it would help forces comply with legal and retention requirements and manage images consistently.

When the NPCC discussed the initiative with Computer Weekly in November 2023, it confirmed that a national program had begun the previous month but did not name it. A spokesperson said the work would establish a custody-image management regime and review all data then held by UK forces, with implementation over two years.

Legislation and deadlines for reform

The government has been developing a new framework for police biometrics. In July 2025, then-Home Secretary Yvette Cooper told a House of Lords committee that it intended to establish "a proper, clear governance framework" for facial recognition. The Home Office followed in December 2025 with a 10-week consultation covering police biometrics, including live and retrospective facial recognition.

At the consultation's launch, the department said existing laws formed a patchwork that neither gave police confidence to expand use significantly nor consistently reassured the public that the technology would be used responsibly.

The HMICFRS report notes that the King's Speech in May 2026 introduced the Police Reform Bill, including a new legal framework for facial recognition. Subject to parliamentary approval, the bill will apply the same retention and disposal standards to custody images as to DNA and fingerprints.

In its formal consultation response, the government reaffirmed its intention to legislate. It has not yet committed to specific safeguards beyond a new independent oversight body, which will combine and expand existing oversight functions, and a statutory requirement for public participation in oversight.

HMICFRS made 16 recommendations. It set June 30, 2027, as the deadline for chief constables and the NCA director general to ensure that their organizations complete and record data protection impact assessments covering RFR. It also called on the Home Office to introduce, by that date, a requirement for a statutory code of practice governing law enforcement's use of RFR.

By December 31, 2027, the inspectorate said, the Home Office should provide a national facial recognition system supporting consistent, lawful and effective RFR use across England and Wales. It called on the government to establish an independent oversight framework for police facial recognition, including RFR, by the same deadline.

United Kingdom

Share

Twitter/XFacebookLinkedIn

Read also