Lighthouse AdvisorySLED AI Adoption Intelligence
← Back to results

From the Public Safety edition of October 31, 2026

Independent researchCautionaryPreprint, not peer reviewedNew this fortnight

Independent comparative study finds European supervisors learned of police facial recognition misuse from the press in all four documented cases, and often could not reconstruct use or show corrections were made

AEGIS (independent research; author Tom Anciaux Berner) · Policing / facial recognition oversight · Sweden, Belgium and France

Publisher
Zenodo (AEGIS Research Paper 01)
Original publication
Zenodo, AEGIS Research Paper 01, Expert Review Edition v1.0, 27 Oct. 2026
Source retrieved
2026-11-01
Read original source

What happened

This source-based pilot, released for expert review on Oct. 27, 2026, examines four documented police facial recognition uses: Swedish and Belgian police use of Clearview AI, the Brussels Airport system, and BriefCam’s face function in France. In all four, supervisory action followed a public report, three times press coverage. Belgium’s supervisor relied on Clearview’s own figure of 78 searches, which it could not verify, and found no police logs; an earlier police answer denying use was later contradicted. At Brussels Airport, equipment bought in early 2017 was unknown to the supervisor until a 2019 interview. In France, connection logs were deleted after 365 days, so inspectors relied on declarations. Sweden’s SEK 2.5 million penalty was annulled on appeal. The author says the evidence does not show supervision is ineffective: once informed, supervisors acted, and Belgium halted a deployment.

Why it matters

Directly relevant to U.S. oversight design: it separates whether a control exists, whether it worked and whether that can be demonstrated, and shows how vendor-held logs and short retention defeat audits.

Evidence and measured results

Downloaded and read the full PDF from Zenodo, including methodology, the four case sections and stated limitations. Primary regulator decisions cited in the paper were not opened.

Limitations and uncertainty

Preprint, not peer reviewed (expert review edition; open items listed by the author). Four publicly documented European cases, not a representative sample; events 2017-2024 under EU-era law; transferability to U.S. agencies is limited by different legal frameworks.

Put this evidence to work

Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-01; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.

Sales

Role takeaway
Customer problem
Oversight bodies cannot independently show how police facial recognition was used or that corrections were made.
Stakeholders
  • Inspector general or civilian oversight board
  • Police compliance unit
  • City or county attorney
  • Records and IT security
Discovery questions
  • Where are search logs held and for how long?
  • Could an auditor reconstruct use without vendor input?
  • How are free trials and vendor-added features detected?
Value hypothesis
An auditability assessment could show oversight bodies where evidence would be missing before a controversy arises.
Potential engagement
Facial recognition and analytics auditability review for one agency.
Unsupported claims
  • That supervision is ineffective in general
  • Prevalence of unlawful use
  • U.S. legal conclusions from EU cases

Pre-sales engineering

Role takeaway
Fit
Relevant to agencies using face search or video analytics, and to their overseers.
Architecture and integration
Usage logs across vendor platforms and agency systems; retention and access controls determine audit reach.
Prerequisites
  • Inventory of tools and features
  • Agency-held usage logs
  • Retention schedule aligned to audits
Constraints
  • State retention laws vary
  • Vendor contracts may hold logs
  • Study is European and pre-AI Act
Security
Ensure logs are tamper-evident and readable by auditors without vendor mediation.
Proof of value
  • Reconstruct three months of use from agency-held logs and compare with vendor counts.

Delivery

Role takeaway
Work
Build a tool register, align log retention, give auditors access, track corrective actions to closure.
Dependencies
  • Vendor cooperation on log export
  • Oversight body mandate
  • IT security support
Ownership
Agency compliance or professional standards unit
Skills and adoption
Train supervisors to recognise and report unsanctioned tool use.
Governance checkpoints
  • Annual tool register review
  • Auditor access test
  • Closure evidence for findings
Acceptance criteria
  • 100% of AI tools and trials registered
  • Logs retained at least one audit cycle
  • Every corrective order closed with evidence
Risks
  • Shadow trials outside the register
  • Vendor-held logs unavailable
  • Retention shorter than audit cycle

Implementation considerations

Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.

Architecture and integration

What must connect, and where does the AI sit in the workflow?

Audit reconstruction depended on log retention and on whether usage records were held by the police or the vendor.

Governance

Who approves, reviews and stays accountable for outcomes?

Supervisors depended on information supplied by the supervised agency or vendor; proof that corrective orders were carried out was often absent from the public record.

Security and privacy

What data, permissions and controls need testing?

Uploading police photos to an external commercial service was found to be a data breach in Belgium; deletion could not be confirmed publicly.

Accessibility and workforce

Who is affected, and what skills or accommodations follow?

Individual officers used trial tools outside sanctioned channels, which organisational controls did not catch.

Procurement

What should contracts, pricing and exit terms secure?

Free trial licences and vendor-added features (BriefCam’s face function) entered operations without procurement review.

Operating model

Which teams own the service once it runs?

Oversight bodies need independent log access and verification of corrective actions, not only legal powers.

Publication history

  1. 2026-10-31Public Safety · Issue 556 resources
Read preserved resource versions (JSON)

Stable resource ID: aegis-police-facial-recognition-misuse-oversight-sweden-belgium-france-20261027