Over 100 people charged after South Wales AFR Identify searches

Cardiff University's evaluation relays a count of over 100 people charged after AFR Identify searches. With the newer algorithm, operators confirmed matches in 22% of 1,524 searches.

  • 2 for an organization

    Operators confirmed system matches in 333 of 1,524 searches with the newer algorithm, and the report relays a count of over 100 people charged, with no comparison against investigations without the system.

  • Not rated society

    The report gives no outcome for the charges and no count of crimes solved or prevented.

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

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

Only the evaluators support the benefit.

Size

No magnitude yet: distribution is not documented.

The system’s part
Not isolated. The evaluators counted operator-confirmed matches and compared the two algorithm versions, but they did not compare investigations with and without the system. Operators decide each match, and the quality of the submitted image shapes the result. The report says that NEC has not revealed how its algorithms calculate matches, and it does not say whether they learn from data.
Learns from data

Shown. NEC's NeoFace Watch brochure says that its face recognition relies on a modified Generalized Learning Vector Quantization algorithm, a supervised learning method. The brochure does not give the version that the force used, and the evaluation says that NEC has not revealed how its algorithms calculate matches.

“NEC's face recognition relies on a modified Generalized Learning Vector Quantization (GLVQ) algorithm.” Source

Later
Persisted. The report counts outcomes to May 2018 as 675 possible matches to suspects, over 350 of which led to charging decisions.
Not yet checked
LimitsCosts and harms

Download data Cite Suggest a correction

Benefit type
Reduced Violence
System
AFR Identify, using NEC's NeoFace facial recognition
Made by
NEC
Used by
South Wales Police
Where
South Wales, United Kingdom
Stage
In a pilot
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Physical health (crime reduction or prevention)
Collections
Cross-border benefits
Last revised
1 November 2026

Sources

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

    Davies, Bethan, Martin Innes and Andrew Dawson. An Evaluation of South Wales Police's Use of Automated Facial Recognition. Universities' Police Science Institute, Cardiff University, September 2018. Wayback Machine copy of https://www.cardiff.ac.uk/__data/assets/pdf_file/0006/2426604/AFRReportDigital.pdf.

    Over 100 people charged following use of AFR Identify

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

    NEC. NeoFace Watch: High performance face recognition (product brochure), copy hosted by naffco.com.

    NEC's face recognition relies on a modified Generalized Learning Vector Quantization (GLVQ) algorithm.

Details

AI system

Reported configuration

NEC S17 algorithm until October 2017, then the M20 algorithm

Workflow

Staff in the Identify team searched still images of unidentified suspects against a custody database of about 450,000 people. The system returned up to 200 results ranked by similarity score, and an operator chose any that might be a match. NEC supplied an updated algorithm in October 2017, and the force improved how it took custody images.

Follow-up

Conditions for replication

Submitted images must be of good enough quality, because the system rejected poor ones. An operator must review each possible match, since the system only suggests matches. Officers must treat results as intelligence that supports their own inquiries.

Entry details

Sector

Policing (suspect identification).

Purpose

NEC built NeoFace to compare a face in an image with a database of stored images and return possible matches with similarity scores. South Wales Police used its Identify mode to generate investigative leads from images of unidentified suspects.

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 benefit

Peter H. Vartanian. (2026). Over 100 people charged after South Wales AFR Identify searches (AIBD-1219). Artificial Intelligence Benefits Database, edition of 2 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1219.html (archived at https://doi.org/10.7910/DVN/AADSII, version 9.0)

BibTeX Data Field definitions