As Facial Recognition Technology Expands, UK Explores ‘Inferential’ Surveillance Systems

British police agencies are exploring a new class of so-called “inferential” surveillance systems that attempt to interpret behaviour, stress, and even emotional cues.

On Dec. 4, 2025, ministers launched a public consultation to create a legal framework on facial recognition inferential technologies, saying they could help police catch criminals, prevent suicides, and locate missing people.

There is currently no dedicated statute governing police use of facial recognition in England and Wales.

Forces rely instead on common law powers alongside data protection and human rights legislation, stating that deployments must serve a legitimate policing purpose and be necessary, proportionate, and fair.

Supporters of expansion point to operational results. The Metropolitan Police reported 962 arrests from live facial recognition deployments between September 2024 and September 2025, figures often cited by ministers as justification for wider rollout.

UK police agencies actively deploy systems that scan live video feeds in public places and compare faces against watchlists of wanted individuals.

According to a recent government factsheet, police now use three forms of facial recognition: Retrospective Facial Recognition, Live Facial Recognition, and Operator-Initiated Facial Recognition.

Retrospective Facial Recognition is used after incidents as part of criminal investigations, drawing on images from CCTV, mobile phones, dashcams, doorbells, and social media. These images are compared against custody photographs held on the Police National Database. Forces carry out more than 25,000 facial image searches each month, with some also operating local Retrospective Facial Recognition systems.

The National Police Chiefs’ Council reported that 127 suspects who engaged in serious disorder across the UK in August 2024, were identified using Retrospective Facial Recognition following protests sparked by the murder of three girls who were killed in a knife attack at a Taylor Swift-themed holiday club by Axel Muganwa Rudakubana.

Live Facial Recognition compares live footage of passersby against watchlists in real time. Authorities said deployments in London between January 2024 and September 2025 led to more than 1,300 arrests for offenses including rape, domestic abuse, aggravated burglary, grievous bodily harm, robbery, drug supply, animal cruelty, and offenses against children.

Police also said the technology helped ensure compliance by registered sex offenders, leading to more than 100 arrests.

As of November 2025, 13 police forces have used or are using Live Facial Recognition, including the Metropolitan Police, South Wales Police, Greater Manchester Police, West Yorkshire Police, Surrey, Sussex, Essex, Bedfordshire, Suffolk, Hampshire and Isle of Wight, Leicestershire, Northamptonshire, and North Wales.

Operator-Initiated Facial Recognition allows officers to photograph a person of interest using a mobile app to confirm identity without making an arrest. South Wales Police and Gwent Police are currently using the system.

Surveillance in Societies

Eleanor ‘Nell’ Watson, a leading researcher and adviser on artificial intelligence ethics and transparency, has criticized the increased deployment of surveillance technology.

“The UK is constructing infrastructure for a surveillance society while telling itself it is merely catching criminals,” she told The Epoch Times via email.

Watson holds the position of executive consultant on philosophical matters for Apple, as well as the president of EURAIO, the European Responsible Artificial Intelligence Office. She also serves as senior scientific advisor to The Future Society, and senior fellow at The Atlantic Council.

She said that the UK began deploying CCTV extensively in the 1990s following IRA bombings, creating both physical infrastructure and institutional comfort with public surveillance that continental Europe never developed.

“London now has approximately 68 CCTV cameras per 1,000 people, roughly six times Berlin’s density. This pre-existing network makes adding inferential capabilities feel like an incremental upgrade rather than a paradigm shift,” Watson said.

She said that facial recognition works by extracting geometric measurements from facial images, the distance between eyes, the shape of the jawline, the relative position of facial landmarks, and converting these into mathematical representations called “faceprints” or “embeddings.”

“These are compared against databases of known individuals using similarity algorithms. Modern systems use deep neural networks trained on millions of facial images to improve accuracy,” she added.

For standard identification purposes, matching a face against a watchlist of known suspects, the technology has become reasonably reliable under controlled conditions.

The Metropolitan Police reports a false positive rate of roughly 1 in 1,000 for their current live facial recognition deployments.

Watson said that this apparent accuracy masks significant problems: accuracy varies substantially by lighting conditions, camera angle, and critically, by demographic group. Multiple studies have documented higher error rates for darker-skinned individuals and women.

She added that inferential technologies, systems attempting to detect emotions, stress, deception, or “suspicious behaviour,” rest on “far shakier scientific foundations.”

“The core assumption underlying emotion recognition is that internal emotional states produce reliable, universal external expressions,” she said, noting that this assumption has been comprehensively challenged.

A landmark 2019 meta analysis published in Psychological Science in the Public Interest, authored by five researchers who began the project with different theoretical commitments, concluded that facial expressions are not “fingerprints” or “diagnostic displays that reliably and specifically signal particular emotional states regardless of context, person, and culture.”

“Put simply: a frown does not reliably indicate anger. A smile does not reliably indicate happiness. Emotional expressions vary enormously across cultures, contexts, and individuals,” Watson said.

She said that her assessment is that the UK is constructing what might be called “surveillance infrastructure with democratic characteristics,” building capabilities comparable to authoritarian systems while maintaining the procedural forms of liberal democracy.

“The risk is that infrastructure, once built, shapes the political possibilities available to future governments. A system capable of comprehensive behavioural monitoring does not become less capable because a friendly government is in power,” she said.

The UK Home Office did not respond to The Epoch Times request for a response.

Inferential Surveillance

Demetrius Floudas, former policy and geopolitical adviser to Cabinet-level decision makers, including in the British Foreign Office, said “Inferential technologies can be a profound intrusion, akin to mind-reading by algorithm.”

He said the “first plausible challenge” to the use of the technology would likely be a judicial review arguing that inferential surveillance violates Article 8 of the European Convention on Human Rights (ECHR), which guarantees the right to respect for private and family life, one’s home, and correspondence, and possibly Article 6, which guarantees the right to a fair and public hearing within a reasonable time.

“With Britain already possessing a very dense network of CCTV, adding inferential surveillance is simply the next logical step in monitoring public spaces,” Floudaas told The Epoch Times via email.

“This is coupled with political appetite for being a global leader in security tech post-Brexit, a constant increase in crime rates and a completely overwhelmed police force that has been forced to ignore all non-serious offences for several years now.”

Nora Demleitner, a prominent U.S. criminal-justice scholar, former law-school dean, and elected member of major legal think tanks, said the UK is already further ahead than the United States in the police surveillance area.

Demleitner said that with a large amount of footage available, from a law enforcement perspective “it makes sense to want to include more facial recognition technology.”

She said she expects a governmental push to capture more faces to enlarge the databases and to improve the quality and number of public cameras, which will move toward a greater surveillance architecture.

“We see lots of cross-fertilization in policing between the US and the UK and should expect the models to influence each other. The UK is farther along on a more broad-based surveillance model,” she told The Epoch Times via email.

U.S. federal prison consultant and criminal justice advocate Christopher Zoukis said the United States has no federal ban on police facial recognition, with use governed at state and city level, allowing real-time deployments in some jurisdictions while others have imposed outright local bans.

“The UK’s pivot towards enhanced facial recognition and inferential surveillance is an example of a model centered on public-space safety and rapid technological adoption,” Zoukis told The Epoch Times by email.

“Unlike in the US, where constitutional baseline (the Fourth Amendment) sets criteria for search and seizure and demands that courts look at privacy expectations, the UK law is based on oversight policy; therefore, pilots and deployments can move more quickly,” he said.

“Inferential systems are also more than face identification,” he said.

“They analyze body language, stress signs and behavior profiles to identify issues such as potential criminal activity or indicators of self-harm. These systems depend heavily on data and interpretation, meaning any productive use of them has to involve human review.”

Elizabeth Melton of U.S civil liberties group Banish Big Brother said concerns about emotion-reading surveillance were being raised as early as 2019.

She pointed to claims made by UK firm WeSee, which has said its AI can detect suspicious behavior by reading facial cues imperceptible to the untrained eye, according to a BBC report.

“Imagine walking through an airport after a painful breakup, or your dog has just died, and your distress is now construed as dangerous,” Melton told The Epoch Times by email.

“Or someone fears that you might be suicidal and must therefore intervene. This approach threatens to rob us of the freedom to express the very emotions that make us human.”

She also warned that such systems could disproportionately affect people who express emotions differently, including individuals with autism.

Historically, she said, people living under authoritarian regimes have learned to tightly regulate how they speak, act, and even dress to avoid attracting attention.

“That is not the kind of atmosphere that should characterize a free and open society,” she said.

Laws in Europe

In EU member states, the EU AI Act, in force since August 2024, places strict limits on law-enforcement use of biometric and behavioral AI, including tight necessity and proportionality tests and high-risk classifications for any deployment.

In France real-time public use of Facial recognition is generally banned under current law, though political debate is underway about changing or updating legislation on the issue.

Germany has strong constitutional privacy protections. But police in Germany are employing high-definition cameras and live facial recognition to catch suspects, raising legal questions in the wake of the European Union’s AI Act.

The system, which is deployed by police in the German eastern State of Saxony and in Berlin, can process facial images “with a time delay of a few seconds,” the Berlin public prosecutor’s office told German media outlet Netzpolitik last year.

In Spain, police use of facial recognition is highly constrained, as any project involving this technology requires a Data Protection Impact Assessment (DPIA), an evaluation that examines potential risks to the rights and freedoms of affected individuals.

Italy is among the most restrictive major EU states. The Italian data-protection authority (Garante) has repeatedly intervened against facial-recognition deployments, effectively blocking or suspending real-time public use.