High False Positive Rate by SWP's Facial Recognition Use at Champion's League Final

South Wales Police (SWP)’s automated facial recognition (AFR) at the Champion's League Final football game in Cardiff wrongly identified innocent people as potential matches at an extremely high false positive rate of more than 90%.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The primary failure was the facial recognition system's extremely high false positive rate of 92%, demonstrating a severe lack of capability and robustness under real-world deployment conditions.

Additional risk subdomains

• 1.3 Unequal performance across groups: The Court of Appeal found that the police failed to comply with the Public Sector Equality Duty by not verifying if the software possessed racial or gender bias.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by the NeoFace Watch AI system generating an extremely high rate of false positive matches post-deployment, which was an unintentional outcome of its security monitoring goal.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The system is a law enforcement application using remote biometric identification (facial recognition) in public spaces, which is explicitly classified as high risk under the EU AI Act.

AI system and alleged parties

• AI system: NeoFace Watch (NEC)
• AI purpose: Face Recognition; Identification
• Behaviour type: Tool
• Alleged developer: NEC
• Alleged deployer: South Wales Police
• Alleged harmed parties: FInals attendees, falsely accused Finals attendees

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Minor, indirect Negligible
• Civil rights: direct Substantial, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Substantial, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Differential treatment

Reported: Yes

Directly caused: The system falsely flagged 2,297 innocent individuals as potential criminals, subjecting them to potential police intervention and questioning that other attendees did not face.

Inferred additional harm: There is a risk of indirect discrimination on the basis of sex or race, as noted by the Court of Appeal, because the police failed to verify if the software had demographic biases.

Civil rights

Reported: Yes

Directly caused: The Court of Appeal ruled that the use of the facial recognition technology was unlawful and violated Article 8 of the European Convention on Human Rights (the right to respect for private life) for the individuals scanned.

Inferred additional harm: The unlawful deployment of real-time surveillance on 170,000 citizens represents a broad interference with civil liberties and privacy rights.

Privacy

Reported: Yes

Directly caused: The deployment violated the privacy rights of the scanned individuals under Article 8 of the ECHR and failed to comply with the Data Protection Act 2018, specifically regarding the processing of biometric data of 170,000 people.

Inferred additional harm: The biometric data of thousands of innocent citizens was processed without their explicit consent or a sufficient Data Protection Impact Assessment.

Psychological

Reported: Yes

Indirectly caused: The false positive matches led to police interventions where officers had 'chats' with innocent individuals to verify their identity, likely causing temporary distress, confusion, and anxiety.

Inferred additional harm: It is likely that many of the 2,297 falsely matched individuals experienced mild to moderate distress or embarrassment during police interactions, though specific numbers are not documented.

People affected

• Occurrences reported: 1
• People reportedly harmed: 2297
• People reportedly exposed: 170000

Potential causes

Management

• Lack of Transparency: Management kept demographic data and internal audits hidden from the public.
• Insufficient Risk Oversight: Failed to establish redress mechanisms for individuals harmed by the tech.

Technology

• High False Positive Rates: The NeoFace system had a 92% false positive rate at the Champions League.
• Unknown Training Data Bias: Police did not know the training dataset, risking demographic bias.

Data Inputs

• Poor Quality Watchlist Images: Low-quality photos from partner agencies like UEFA caused matching errors.
• Blanket Biometric Processing: System scanned faces indiscriminately, processing data of innocent people.

Human Factors

• Broad Officer Discretion: Officers had wide discretion on watchlist inclusion and deployment locations.

Process and Methods

• Deficient Impact Assessments: DPIA failed to assess risks to rights and potential demographic bias.
• No Bias Verification Process: No process was in place to test or verify the software for racial or sex bias.

Regulatory Environment

• Inadequate Legal Framework: Lack of clear legal standards governing the deployment of live facial scans.
• Non-compliance with PSED: Failure to meet Public Sector Equality Duty requirements before deployment.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide highly detailed, consistent factual accounts of the incident, including specific statistics on false positives (2,297 out of 2,470), the specific AI software used (NEC NeoFace Watch), and the detailed legal findings from the Court of Appeal ruling.
• Ambiguities identified: None of significance; the technical performance metrics and legal outcomes are clearly documented across multiple reputable sources.
• Alternative interpretations: None; the facts of the deployment, the high error rate, and the subsequent legal ruling are uncontested.

South Wales Police's unlawful deployment of NEC's facial recognition system at a major sporting event resulted in a 92 percent false positive rate, violating privacy rights for thousands. The subsequent landmark legal ruling established key precedents regarding the human rights and data protection boundaries of public biometric surveillance.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: United Kingdom
• Alleged perpetrator: South Wales Police

Threat characteristics

• Imminence: Long-term. This reflects a past deployment and subsequent judicial ruling, representing long-term policy and regulatory concerns rather than an active crisis.
• Autonomy: Human-supervised. The AI system autonomously generated match alerts, but human police officers retained final decision-making power regarding interventions.
• Novelty: Evolved capability. The deployment scaled up real-time public biometric surveillance, representing a significant evolution of existing automated facial recognition capabilities.

Impact by dimension

• Physical security: Negligible. The incident involved public surveillance and did not pose any threat to physical systems, critical infrastructure, or kinetic safety.
• Information security: Negligible. No intelligence compromise, classified data theft, or hostile information warfare operations were associated with this deployment.
• Sovereignty: Minor. While a domestic government agency deployed an unlawful tool, the issue was successfully resolved through established judicial and constitutional processes.
• Economic security: Negligible. The deployment of the facial recognition tool did not impact strategic industries, critical supply chains, or financial systems.
• Societal stability: Substantial. The unlawful real-time biometric scanning of 170,000 citizens violated privacy rights under Article 8 of the ECHR and failed to address potential demographic biases.