All research
Consultation response··Updated ·6 min read·Facial Recognition·Civil Liberties·UK Policy

Facial Recognition in the UK: Built Before Parliament Authorised It

Police facial recognition is being bought and deployed before there is a specific law for it. A summary of our response to the Home Office consultation, with sources.

On 11 February 2026 we submitted a response to the Home Office consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies (deposited in Parliament as DEP2025-0828). This is a condensed version of the argument. It was revised in November 2026 to add sources and correct several figures.

Our view of the law is this:

Mass biometric surveillance of public spaces without judicial pre-authorisation, statutory basis, or adequate equality protections is incompatible with ECHR Articles 8, 10, 11, and 14, Data Protection Act 2018 Part 3 principles of strict necessity and proportionality, and the Public Sector Equality Duty under the Equality Act 2010 s.149.

Deployment came before the law

There is no legislation specifically on police facial recognition. The consultation paper describes the legal basis as “a patchwork of common law, data protection, and human rights legislation”. Meanwhile:

  • The government funded ten live facial recognition vans in 2025 and announced 40 more in its January 2026 policing white paper (paragraph 296).
  • The Metropolitan Police has mounted live facial recognition cameras on street furniture in Croydon (Met news release).
  • Police run more than 25,000 retrospective facial recognition searches a month on the Police National Database, according to the consultation paper.

All of this was in place or announced before the consultation closed. The infrastructure came first and the law has not yet followed.

The bias data is public

The National Physical Laboratory (NPL) tested the algorithm used for those Police National Database searches, Cognitec FaceVACS-DBScan ID v5.5. Its report, dated October 2025 and published in December 2025, found that at a match threshold of 0.8 the false positive identification rate was:

  • 0.04% for white subjects
  • 4.0% for Asian subjects
  • 5.5% for Black subjects
  • 9.9% for Black women

The rate for Black subjects is about 138 times the rate for white subjects. At a threshold of 0.9, NPL recorded no false positives for any group.

So the threshold matters. Liberty Investigates and the Guardian reported in December 2025 that the National Police Chiefs’ Council, told of the bias in September 2024, raised the threshold to reduce it, then reversed the decision the following month after forces complained that the system was producing fewer leads. According to that report, the higher threshold cut the share of searches returning a potential match from 56% to 14%.

NPL also tested an Idemia algorithm that it says will be used by the Home Office Strategic Facial Matching service. At face-match thresholds between 3052 and 3981, which NPL calls the optimum range, its report found no false positives and no demographic variation.

The bias is documented and a better-performing algorithm has been tested. Running the current algorithm at a threshold where the error rates differ by ethnicity and sex is a choice, and it engages the Public Sector Equality Duty under section 149 of the Equality Act 2010.

In our view the proposals fell short in four ways.

Wrong statutory home. Live facial recognition by police is law-enforcement processing, governed by Part 3 of the Data Protection Act 2018 and not by the UK GDPR. Part 3 sets stricter necessity and proportionality tests. The consultation papers did not ground their analysis in it.

No PACE-equivalent threshold for stops triggered by an alert. An algorithmic alert is not “reasonable suspicion” under the Police and Criminal Evidence Act 1984. Treating it as such removes a basic protection of English criminal procedure.

No enforceable technical assurance. The proposals did not require vendor transparency, privacy-by-design controls, model attestation or independent audit access.

No reviewable remedy. When the system produces a false match, there is no time-limited route for the individual to correction or compensation.

What we asked the government to do

We made eight requests: four we regard as essential, and four safeguards if expansion goes ahead.

Essential:

  1. Halt expansion until Parliament has legislated specifically on facial recognition.
  2. Withdraw the current algorithm and use the replacement NPL has tested. Restore the higher threshold that was reversed.
  3. Pre-authorisation by a senior judge for any live facial recognition deployment, with published reasons.
  4. No retrospective searches of the passport database without a specific PACE threshold and a judicial warrant.

If expansion goes ahead:

  1. A statutory definition of “deployment” that includes all retrospective database searches.
  2. Independent testing to NPL’s standard every six months, with public results and a limit on demographic error differences.
  3. Custody-record fields recording whether facial recognition triggered the stop, the match score, and outcomes broken down by demographic group.
  4. A two-year sunset clause on any framework legislation, with parliamentary review.

The Home Office consultation asked whether police should be able to search other government databases, such as the passport and immigration databases. The Cabinet Office’s 2026 digital identity consultation (CP 1498) proposed a digital ID built on existing infrastructure including the passport service. Without a statutory firewall, identity verification and police biometric search become two uses of the same records.

Our digital ID consultation response asks for that firewall: no national identity database should be used to train, seed or query a facial recognition system without specific judicial pre-authorisation.

How we handled evidence

In this article, claims taken from government documents are stated directly and linked. Claims that rest on press reporting say so in the sentence.

What happened next

The consultation closed on 12 February 2026. The government published its response in November 2026. It says the government intends to introduce a new statutory framework and a new independent oversight body. As of November 2026 that legislation has not been passed.


The full submission was filed with the Home Office on 11 February 2026. Cosmo Codex is the trading name of CosmoCodex Ltd.

Sources

  1. Home Office, Consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies, December 2025. Source for the “patchwork” description, the 25,000 monthly searches and the passport and driving licence questions.
  2. National Physical Laboratory, Facial Recognition Technologies: Accuracy and Equitability Evaluation of Cognitec FaceVACS-DBScan ID v5.5 (Report 4153), October 2025, section 1.2.2 and Table 5.
  3. National Physical Laboratory, Accuracy and Equitability Evaluation of Idemia Facial Recognition Algorithm for Home Office Strategic Facial Matching (Report 4674), October 2025, section 1.2.
  4. Liberty Investigates with the Guardian, UK police forces lobbied to use biased facial recognition technology, 10 December 2025. Press reporting based on NPCC documents.
  5. Home Office, From Local to National: A New Model for Policing (CP 1489), January 2026, paragraph 296.
  6. Home Office, Live Facial Recognition technology to catch high-harm offenders, 2025.
  7. Metropolitan Police, 100 arrests following new Live Facial Recognition pilot in Croydon.
  8. Equality Act 2010, section 149.
  9. Home Office, Government response to the consultation, November 2026.

Interested in working with us?

Get in touch to discuss our products, partnerships, or research collaborations.