Police facial recognition use in the UK must be limited to “strictly necessary and proportionate” cases to protect people’s collective and individual rights, says civil society coalition.
Initially announced by then-home secretary Yvette Cooper in July 2025, the UK government is set to establish a specific legal framework for police facial recognition, which the King’s Speech in May 2026 confirmed would be contained within the upcoming Police Reform Bill.
According to the King’s Speech, the framework will make clear “when use of these technologies can be justified” and includes the “creation of a single, expert regulatory body to provide independent advice and oversight”.
Responding to the announcements, 12 digital rights groups – including Big Brother Watch, Liberty, Justice, the End Violence Against Women Coalition and the Open Rights Group – have signed a joint statement mapping out the “minimum, necessary protections” needed to ensure police facial recognition does not trample over key human rights.
This includes protections against both live facial recognition (LFR), which acts like a biometric checkpoint by scanning the faces of passers-by in real time, and retrospective facial recognition (RFR), which is used by police to analyse recorded footage. Both are matched against a watchlist to identify individuals of interest to the police.
The joint statement outlined the “minimum, necessary protections” needed to protect human rights, and urged the Home Office to prohibit the use of LFR except for “strictly defined policing purposes”.
This includes where there is a threat to life, reasonable grounds to believe the person has committed a serious criminal offence, and in the targeted search for missing persons and victims of kidnapping or trafficking.
The proposed framework aims to ensure that LFR and RFR are used for targeted and justified reasons, rather than as an everyday policing tool, to protect the public’s freedom to privacy and expression.
“They represent the minimum standard the bill must meet in order to provide a clear, coherent and sustainable framework for facial recognition that mirrors our democratic counterparts [in Europe], whilst offering some protection to the general public from excessive AI surveillance,” wrote the groups.
LFR vs RFR requirements
On LFR deployments, the coalition recommended requiring detailed justifications by law enforcement bodies, including that the “individual sought person on the watchlist will be present at the proposed deployment location”, and that police should have exhausted all other measures of locating the person in question.
LFR is also recommended not to be used in places with “greater expectations of privacy”, such as schools, places of worship, health centres, polling stations or protests, to protect privacy and freedoms of expression.
Each deployment should also be authorised by a judicial or independent administrative authority to ensure its use is just.
If urgent deployment is necessary, documentation is further required to justify its use in proportion to “human rights interference” – including with a human rights impact assessment, geographical relevance, if less intrusive methods have been exhausted, and its impact on vulnerable groups.
Documentation, such as human rights impact assessments, would also need to be completed by the deploying law enforcement body and submitted to an authorising body.
The recommendations for RFR are similar to those for LFR, where the sought person must only be matched to images obtained lawfully, that are up to date, to prevent wrongful identification – this does not include composite images such as police sketches.
As it stands, however, the Home Office continues to hold millions of custody images unlawfully.
With RFR, the groups have called for police to be prohibited from using image references taken from databases created via the untargeted scraping of facial images, or databases that include people who are not under investigation, without criminal convictions, or those comprised of non-police images.
Both LFR and RFR recommendations include a requirement to publish outcomes of their deployments – such as the number of uses, outcomes of matches, if it resulted in arrests, convictions or wrongful identification, and demographic data.
The groups also envisage the establishment of an independent oversight body, which is empowered to conduct post facto reviews of both individual LFR deployments and their systemic use, set scientific validation requirements around accuracy and reliability, set various standards around design and performance, and conduct in-depth audits of police systems.
As it stands, the technology’s use is not regulated by any specific regulatory framework and instead operates via a “patchwork” of other laws, including data protection rules and common law policing powers.
In December 2025, the Home Office launched a 10-week consultation on a legal framework.
While it is yet to formally respond to the consultation, the process aimed to gather views on establishing a framework, such as whether the technology should be designed to access and match facial imagery from other existing databases – including passport and driving licence images, custody records, and immigration files.
Beyond LFR, public opinion was consulted on other “biometric and inferential technologies”, such as emotion detection to identify criminal or self-harm behaviour, voice and gait recognition.
At the time of the consultation’s announcement, the Home Office described the current rules in place for LFR as “complicated and difficult to understand”, with Cooper adding that the new framework was necessary to give police forces the “confidence to use it with the right standards in place”.
Despite this, and before the consultation had even closed, the Home Office announced sweeping reform to UK policing in January 2026, which included significant investment in the LFR capabilities of police in England and Wales.
Previous legal challenge
The move for “significant reforms to the police” outlined in the King’s Speech came after a landmark legal challenge against the Metropolitan Police’s use of LFR was dismissed by the UK High Court on 21 April 2026, where it was ruled that “the [Met’s LFR] Policy contains adequate and lawful constraints”.
Met commissioner Mark Rowley responded to the ruling, saying that the judgment confirmed that the Met was “deploying this technology responsibly and with care”.
Brought forward by Big Brother Watch’s director Silkie Carlo and anti-knife campaigner Shaun Thompson, who was incorrectly identified by LFR technology and described LFR as “stop and search on steroids”.
The legal challenge cited concerns with discrimination and the disproportionate use of LFR in areas with large ethnic minority groups, as well as issues around the permissiveness of the Met’s deployment criteria.
Discriminatory algorithms
Accuracy and bias risks were found in Essex Police’s use of LFR in March 2026, leading to the technology’s suspension by the force. Computer Weekly exclusively reported in 2025 that Essex Police had failed to consider the technology’s potential discriminatory impact.
Because LFR software generates a similarity score when a match is identified, which triggers an alert at a set threshold, law firm TSABI argued in a June 2026 blog post that if this threshold is set too low, “especially in crowded or poorly lit public order scenarios where the software is most likely to make mistakes”, it risks leading to wrongful identification. TSABI added that certain algorithms perform less accurately on Black and Asian faces.
Similarly, a Cambridge study from March 2026 found that the technology correctly identified men over women and Black participants over other ethnic groups.
Despite its reported bias and lack of regulatory frameworks, the government is pushing ahead to roll out LFR technology.
Currently, 13 of 43 police forces across the UK have deployed LFR technology, with Statewatch reporting that four additional regional police forces across England used the technology for the first time in the past six months.
