Full Report
Cams statistically more likely to ID Black people, says new research A UK police force has suspended its deployment of live facial recognition (LFR) technology after a study revealed it was statistically more likely to identify Black people on a watchlist database.…
Analysis Summary
# Regulation/Compliance: UK Public Sector Equality Duty (PSED) & Live Facial Recognition (LFR) Operational Standards
## Overview
Recent research involving Essex Police has highlighted significant compliance risks regarding the use of Live Facial Recognition (LFR). Under UK law, public authorities must ensure that technological deployments do not result in unlawful discrimination. This specific case centers on "statistical bias" where LFR algorithms demonstrated a higher rate of false-positive identifications for Black individuals, triggering a suspension of the technology to ensure alignment with human rights and equality mandates.
## Key Details
- **Issuing Authority:** UK Home Office, College of Policing, and the Equality and Human Rights Commission (EHRC).
- **Effective Date:** PSED is currently in effect (Equality Act 2010); specific LFR guidance is evolving.
- **Jurisdiction:** United Kingdom (specifically England and Wales for this deployment).
- **Status:** In Effect (with active enforcement through judicial review and internal police audits).
## Requirements
### Mandatory Requirements
1. **Public Sector Equality Duty (PSED):** Organizations must eliminate unlawful discrimination and foster equality of opportunity.
2. **Data Protection Impact Assessment (DPIA):** Must be conducted to assess risks to privacy and bias before deployment.
3. **Accuracy and Bias Testing:** Must verify that the algorithm does not produce "statistically significant" disparate impacts based on protected characteristics (race, gender).
4. **Legality and Proportionality:** Deployments must be targeted (e.g., specific watchlists) and necessary for a legitimate law enforcement aim.
### Recommended Practices
1. **Independent Academic Auditing:** Commissioning third-party studies (e.g., Cambridge University) to validate system performance.
2. **Software Updates:** Regular patching of biometric algorithms to the latest versions to mitigate known demographic differentials.
3. **Human-in-the-Loop:** All automated alerts must be reviewed by a human officer before intervention.
## Affected Organizations
- **Industries:** Law Enforcement, Public Safety, and Government Agencies.
- **Organization Size:** All UK police forces (specifically the 43 territorial forces).
- **Geographic Scope:** United Kingdom.
## Compliance Timeline
- **April 2011:** Equality Act 2010 (PSED) came into force.
- **Ongoing (2024-2026):** Home Office rollout of 40+ LFR-equipped vans.
- **March 2026:** Deployment suspension by Essex Police following the Cambridge study findings.
- **Future:** Mandatory national standards for AI in policing (under current White Paper proposals).
## Implementation Guidance
### Assessment Phase
- Audit existing LFR algorithms for "False Positive Identification Rates" (FPIR) across various demographics.
- Review the composition of the "Watchlist" to ensure it is lawful and current.
### Implementation Phase
- Work with software providers (e.g., Azure AI, NEC, or others) to implement "bias-corrected" facial recognition models.
- Update operational policies to include a protocol for when bias is detected.
### Validation Phase
- Recalibrate the "Operational Setting" (the threshold at which the software triggers an alert) to balance sensitivity and accuracy.
- Conduct follow-up academic assessments to confirm that updates have mitigated the statistical imbalance.
## Technical Requirements
- **Threshold Tuning:** Adjusting similarity scores to reduce false positives in identified high-risk demographic groups.
- **Watchlist Management:** Secure, encrypted databases (pre-configured) limited to suspects, persons of interest, or missing vulnerable people.
- **System Logging:** Detailed logs of every "Match" and "No Match" to facilitate independent auditing.
## Penalties & Enforcement
- **Fines:** Information Commissioner’s Office (ICO) can issue fines under UK GDPR for biased/unfair processing (up to £17.5m or 4% of turnover).
- **Other Consequences:** Suspension of technology use, loss of public trust, and legal challenges via Judicial Review.
- **Enforcement:** The Biometrics and Surveillance Camera Commissioner and the EHRC.
## Related Standards
- **UK GDPR / Data Protection Act 2018:** Governs the "special category" biometric data processed by LFR.
- **Bridges v South Wales Police:** A landmark legal precedent requiring a clear legal framework for LFR use.
- **ISO/IEC 19795:** Standards for biometric performance testing and reporting.
## Resources
- **Official Documentation:** [hXXps://www.gov.uk/government/publications/police-use-of-live-facial-recognition-technology]
- **Guidance Documents:** College of Policing Authorised Professional Practice (APP) on LFR.
- **Tools:** NIST Face Recognition Vendor Test (FRVT) for benchmarking algorithm bias.
## Practical Recommendations
1. **Immediate Action:** If using LFR, cross-reference your algorithm's performance against the Cambridge University findings regarding gender and racial demographic differentials.
2. **Vendor Management:** Demand "Demographic Differential" reports from software providers before renewing contracts.
3. **Policy Review:** Ensure "Pause" protocols are in place should internal or external audits identify a high risk of bias.