Table of contents

What is facial recognition technology?

Facial recognition technology (FRT) is a biometric technology that estimates the degree of similarity between two faces.

UK police forces use FRT to help identify people. FRT identifies people by checking an image of a person against a list of known individuals to find matches. FRT uses artificial intelligence (AI) and machine learning to train systems to recognise faces.

FRT speeds up identification and frees up police time. FRT can be broadly organised into three steps:

  1. Input and facial detection: an image of a face is captured and uploaded.
  2. Features extraction: facial features are extracted and translated into a numerical template.
  3. Classification: the numerical template is used to create a ‘similarity score’ with other numerical templates in a database to verify or identify a person.

Similarity scores indicate the similarity between someone’s numerical template and a numerical template in a database. Types of numerical templates in databases include ID records or people wanted by the police.

Figure 1: Facial Recognition Technologies

Source: Norton security, What is facial recognition and how does it work?

How accurate is facial recognition technology? 

The accuracy of FRT is primarily assessed by the rate of false positives (PDF): how often a facial recognition technology incorrectly matches the identity of a person of interest with another person.  

FRT systems are given minimum thresholds for matching faces; these thresholds are expressed as numerical values between 0 and 1. Thresholds can be adjusted according to the degree of accuracy required, meaning false negatives and false positives can be accounted for. An accuracy threshold of 0.6 is used in some police live facial recognition (LFR) systems. Any lower similarity scores are disregarded.  

Using the 0.6 threshold, 2,067 of 2,077 potential match alerts by Metropolitan Police LFR in 2024–25 were true alerts and led to 962 arrests. The rate of false positives was 0.0003%.  

Throughout 2024, LFR scanned 4.6 million faces across England and Wales and retrospective facial recognition (RFR; see below) was used 252,798 times. 

What types of facial recognition technology are used by UK police forces? 

UK police forces primarily use FRT software from private sector companies including NEC, Cognitec and Idemia. Police use three types of FRT:  

  • Retrospective Facial Recognition (RFR): RFR is used across UK police forces to identify suspects after an incident. Images of a suspect are collected from sources such as CCTV, mobile phones and social media. They are compared with reference images of known individuals in the Police National Database, which holds over 16.5 million facial records. The RFR system lists the most similar images. Matches are verified by an operator and passed to investigating officers. 

Police forces in England and Wales are the largest adopters of FRT, and several of these forces use LFR as well as RFR. Police Scotland and the Police Service of Northern Ireland currently only use RFR. 

How is facial recognition technology regulated and governed? 

The use of FRT is built on common law, existing legislation, codes of practice and police policies. There is no primary legislation specific to FRT or how the police should use it. The College of Policing’s Authorised Professional Practice outlines the procedural guidance. Figure 2 shows the regulatory framework for police use of FRT. 

Figure 2: Regulatory framework for police use of facial recognition technology. 

Source: A Home Office diagram showing the system that governs police FRT

Current regulatory framework for FRT  

In 2020, the Court of Appeal ruled that covert use of automated FRT does not comply with the European Convention on Human Rights. Due to this ruling, police forces must make the public aware when they are using LFR and make sure that their data policies are aligned with the Home Office’s procedures. 

Regional governance of police use of FRT: 

  • is under consideration under the Justice Bill in Northern Ireland 

Other regulatory bodies that may be engaged in governing data and rights protection in relation to FRT are the Equality and Human Rights Commission (EHRC) and the Information Commissioner’s Office. 

Proposed regulatory framework for FRT 

In May 2025, the Ada Lovelace Institute recommended an alternative “centralised approach” to FRT governance, with a “comprehensive legal framework that provides clarity on the limits, lawfulness and proportionality of biometric systems”.  

A government consultation “to help develop a new legal framework for the use of facial recognition” ended in February 2026. It aims to consolidate the current “patchwork” of laws and regulations to make the law easier to understand and help it keep pace with technological development. 

What are the concerns about FRT? 

Research conducted by the Home Office in 2025 found that 64% of the public supported the use of FRT in policing, and that 11% of the public were opposed. Similar results were found in independent research by King’s College London, the Alan Turing Institute, and the Scottish Biometrics Commissioner.  

However, commentators have raised concerns about police use of FRT. Some of these concerns are outlined below.  

Misuse and data protection  

The main concerns about FRT highlighted in the 2025 Home Office report were the risk of FRT misuse and data security. In 2025, the EHRC challenged the Met Police’s use of FRT and highlighted the potential effects on human rights if FRT is misused. 

In 2024, police forces were found to have conducted FRT searches using the passport database, which includes the images of over 45 million people. Some campaigners opposed the use of this database for FRT searches and said it was a “breach of the right to privacy”. 

Cybersecurity is an ongoing concern across sectors. The National Cyber Security Centre handled 204 “nationally significant” cybersecurity incidents in 2024–2025. A 2023 report by the Alan Turing Institute said that police FRT is not invulnerable to cyber threats. 

Proposed solutions  

The government has suggested that data privacy concerns could be addressed by deleting data after a negative match is made using FRT.  

Privacy campaigners have highlighted that people may cover their faces when FRT is being used (except when ‘section 60AA orders’ are in place; these are imposed when police believe that an item is being worn as a disguise).  

Accuracy 

The Home Office research also highlighted public concerns about false identification of suspects. The error rate of LFR has been found to increase by up to 9.3% when used outside a testing environment. Factors influencing the accuracy of LFR include crowded places, poor-quality images or partially concealed faces.  

FRT more broadly is less good at identifying individuals from low-quality and distorted images, such as old or low-resolution custody images and poor-quality CCTV. It has been suggested that using training data with low-quality or partially obstructed images can help to overcome this shortcoming. 

Additionally, a National Physical Laboratory analysis in 2023 showed a “substantial improvement” in the UK’s FRT accuracy compared with analysis from previous years, with no difference in performance between different population groups. 

Bias 

Some analyses have suggested that biases pose significant challenges to the accuracy of FRT, with implications for people’s experiences of the criminal justice system.  

AI models that are used in FRT are trained using large datasets. Racial biases in AI models are well characterised: if the data used to train AI lacks diversity, it can internalise bias in algorithms, which can in turn affect FRT systems used by police forces and have real-world effects. For example, tests on a commonly used RFR algorithm in 2025 showed a higher rate of false positives for faces of Black and Asian people. The RFR technology was implicated in the wrongful arrest of an Asian man in January 2026. 

Research from 2018 suggested that FRT performs better on male faces. This was confirmed in tests on one FRT used by UK police forces in 2025, which found that black women were the subject of the highest percentage (9.9%) of false positive identifications at a 0.8 threshold. 

In 2023, Amnesty International reported that “introducing biased technology into contexts where racial discrimination already occurs will only exacerbate the problem”. In recent years, some have raised concerns about the culture of the Metropolitan Police, which is the largest user of FRT in the UK. In 2023, the Baroness Casey Review into the standards of behaviour and internal culture of the Metropolitan Police identified “institutional homophobia, misogyny and racism, and other forms of discrimination”. 

What is the future of FRT in UK policing? 

Increased use of LFR 

In January 2026, the Home Office announced plans to expand FRT to every regional police force in England and Wales by purchasing 40 new LFR vans. The government says the LFR vans will target violent and sexual offenders in high-crime areas as part of its strategy to “build a safer society for women and girls”. 

The first permanent facial recognition cameras were installed and used with officers present in South London in October 2025. Results from this pilot are not yet published.  

The Home Office’s Immigration Enforcement command is not a police force, but it is due to expand its use of LFR in UK ports that are often used by people trying to avoid detection. Immigration Enforcement aims to identify and locate people who are in breach of a deportation order, or who are wanted for immigration-related offences.  

The National Police Chief’s Council AI Strategy (PDF) says that “ethical deployment” and public engagement are important considerations when “integrating AI into UK policing”. The police aim to use AI to enhance “productivity” and “effectiveness” (PDF) when fighting crime. In January 2026, the Home Office said it would invest “£115 million over the next 3 years to enable the rapid and responsible adoption of AI and automation technologies by the police”, including by creating a National Centre for AI in Policing (‘Police.AI’). It said AI will “free up” 6 million policing hours each year. The Royal United Services Institute reported that Police.AI is integrating AI into LFR.  

Calls for improved oversight  

Although the benefits of AI integration are recognised by regulators and academics, some have stressed the need for improved oversight. 

Some commentators have said that integrating AI into LFR furthers regulatory and ethical challenges. In 2020, the National Physical Laboratory recommended that “decisions should be made by people” when deploying LFR, and on when to engage a member of the public. A report from the UN Interregional Crime and Justice Research Institute and Interpol says that AI systems used in policing should be “built with the functionalities needed to ensure that humans remain in charge during use” (PDF).  

An EU report expressed concerns about reduced human oversight of automated decision-making and said human oversight remains important for spotting errors. As an increase in FRT use coincides with police budget cuts, some campaigners say maintaining human oversight in decisions is important. 

Related links 

Acknowledgments  

Joe Murphy is a Parliamentary Office of Science and Technology (POST) fellow. POST is grateful to the author for kindly giving their time to produce this briefing. 

Questions about this briefing should be referred to Simon Brawley, who acted as parliamentary lead for this work. 

POST would also like to thank the following people for kindly giving up their time during the preparation of this article:  

  • Professor Barry Godfrey, Professor of Social Justice at the University of Liverpool 
  • Professor Ruth Lamont, Thematic Research Lead for Crime and Justice at the House of Commons 
  • Assistant Professor Varuna De Silva, Thematic Research Lead for AI and Digital Technologies at the House of Commons