FaceVACS-DBScan ID ranks the right face first in 3510 of 3513 searches

NPL found that the Cognitec algorithm that police use on the Police National Database ranked the right person first in 3510 of 3513 test searches. False matches varied by ethnicity, gender and age.

  • Not rated an organization

    NPL measured the algorithm on a test dataset, and no outcome of police searches is reported.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Partly reported benefit.
Evidence

foundstudiedreportedclaimed

Only the evaluators support the benefit.

The system’s part
Not isolated. NPL measured the algorithm's own output, before the review by officers that police require. It made no comparison with identification without the algorithm. The report does not say whether the algorithm learns from data. Cognitec writes that its products incorporate deep learning, without naming a version.
Later
Challenged, 4 December 2025. The Home Office reports that it acted on the findings. It procured a new algorithm and had it tested independently, and it asked the police inspectorate, HMICFRS, to conduct an inspection.
Not yet checked
LimitsCosts and harmsLearns from data

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
Cognitec FaceVACS-DBScan ID, the retrospective facial recognition algorithm of the Police National Database
Made by
Cognitec Systems
Used by
Home Office, which provides the Police National Database with the National Police Chiefs' Council, for searches by police forces and law enforcement agencies
Where
United Kingdom
Stage
In routine use
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Identifying Objects, Actions, and Events, Processing Information
Collections
Scrutiny and findings, Cross-border records
Last revised
1 November 2026

Sources

  1. Evaluation report, by an evaluator that the makers or users commissioned.

    National Physical Laboratory (NPL), Data Science and AI Department. Report 4153: Facial Recognition Technologies: Accuracy and Equitability Evaluation of Cognitec FaceVACS-DBScan ID v5.5. Report for the Home Office and the Office of the Policing Chief Scientific Adviser, October 2025, published on GOV.UK on 4 December 2025. Section 1.2.1, Recognition accuracy.

    On this image set, with no thresholding applied, identification performance was nearly perfect (i.e. a TPIR of nearly 100 %). Of the 3513 mated identification searches the top match returned by the algorithm was the correct reference in 3510 cases.
    The FPIR for White subjects (0.04 %) is lower than that for Asian subjects (4.0 %) and Black subjects (5.5 %). • The FPIR for Black Male subjects (0.4 %) is lower than that for Black Female subjects (9.9 %).

  2. Report, by the organization that used it. Context or contrary evidence.

    Home Office. Summary: accuracy and equitability evaluation of Cognitec FaceVACS-DBScan ID v5.5. GOV.UK, 4 December 2025. Section 'What is the Police National Database?'. It gives the scale of use.

    Police use a Cognitec algorithm procured by the Home Office. Forces and law enforcement agencies conduct around 25,000 searches each month.

  3. Other, by the makers. Context or contrary evidence.

    Cognitec Systems. FaceVACS Technology. Product page, accessed 31 October 2026. It describes the technologies in Cognitec's products.

    Cognitec's products incorporate state-of-the-art technologies relevant to working with facial images, in particular deep learning, computer vision, and pattern recognition technologies.

Details

AI system

Reported configuration

V5.5

Workflow

NPL ran 3513 searches for people in the database and 3914 for people who were not, against about 180,000 reference images. The reference images were custody photos from Metropolitan Police Service holdings and a custody-style photo of each of the 401 cohort subjects. The probe images were custody-style, smartphone, ad hoc camera and selfie photos of the cohort subjects.

Follow-up

Conditions for replication

The face-match threshold may need tuning to the quality of the probe images in each use case. Users and investigators must review the search results visually and corroborate any match before they act on it.

Entry details

Sector

Policing.

Purpose

Cognitec Systems built FaceVACS-DBScan ID to compare a face image of an unknown person with a database of reference images and return the closest candidates.

30 more questions not yet checked

model, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). FaceVACS-DBScan ID ranks the right face first in 3510 of 3513 searches (AIBD-1059). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1059.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)

BibTeX Data Field definitions