According to TheRegister.com, London’s Metropolitan Police Service reported that 203 live facial recognition deployments across the capital from September 2024 to September 2025 led to 962 arrests, with cameras triggering 2,077 alerts and only 10 false positives. The arrests included 549 people wanted by courts, 347 individuals police believed might be committing offenses, and 85 people managed by multiple agencies like registered sex offenders. Despite authorities celebrating the technology’s success in removing dangerous offenders, the report revealed that 80% of false positives involved Black individuals, with seven of the eight being Black males. The department defended the performance as within expectations while critics condemned the racial disparities as disturbing.
Sponsored content — provided for informational and promotional purposes.
The Statistical Sleight of Hand in Bias Reporting
The Met’s presentation of false positive rates deserves critical examination. By calculating the rate as 0.0003% based on total faces scanned (3,147,436), they create an artificially reassuring picture. However, when calculated against actual alerts (2,077), the false positive rate becomes 0.48% – a figure that’s 1,600 times higher. This statistical framing matters because in practical policing terms, it’s the alert-based rate that determines how often innocent citizens face police interactions. The department’s claim that demographic imbalances “are not statistically significant” rings hollow when eight out of ten false identifications target a single racial group, particularly given London’s diverse population demographics where Black residents constitute approximately 13.5% of the population according to census data.
The Unsurprising History of Racial Bias in Facial Recognition
These findings continue a troubling pattern well-documented in academic research and previous police deployments. Studies from the National Institute of Standards and Technology have consistently shown that many facial recognition algorithms perform significantly worse on people with darker skin tones, particularly women of color. The technology’s struggle with proper identification of Black individuals isn’t a new revelation – it’s a persistent technical limitation that becomes particularly dangerous when deployed in law enforcement contexts. The Met’s explanation that deployment locations in “crime hotspots” explains the racial disparity essentially admits they’re using biased data to train systems that then reproduce those biases, creating a self-perpetuating cycle of disproportionate policing.
The Accountability Gap in Unregulated Surveillance
As Big Brother Watch correctly notes, no specific legislation governs live facial recognition in the UK, creating a dangerous regulatory vacuum. Police forces are essentially writing their own rules for mass surveillance technology that fundamentally alters the relationship between citizens and the state. The absence of parliamentary oversight means there are no standardized protocols for data retention, independent auditing, or meaningful recourse for those falsely identified. This lack of legal framework becomes particularly concerning when combined with the documented racial disparities – it creates a system where communities of color bear the brunt of experimental technology without democratic consent or legal protection.
The Dangerous Flexibility of Match Thresholds
The report reveals that all ten false positives occurred at the 0.64 match threshold, the highest setting used during the reporting period. This technical detail exposes a critical vulnerability: police can effectively manipulate the balance between catching criminals and harassing innocent citizens by adjusting this single parameter. Lower thresholds would inevitably catch more genuine suspects but would also dramatically increase false positives, particularly among already over-policed communities. There’s no transparent, independent process for determining where this threshold should be set, nor any requirement for police to disclose when they change it. This technical flexibility becomes a policy black box where crucial civil liberties decisions are made without public scrutiny.
What the Support Statistics Don’t Reveal
While the Met highlights that 85% of Londoners support LFR use according to their survey, the demographic breakdown tells a more nuanced story. The strongest opposition comes from LGBT+ communities, mixed ethnicity individuals, and Black respondents – precisely the groups most vulnerable to misidentification and historical over-policing. Younger people (25-34) also show significantly higher resistance than older generations who are least likely to experience the technology’s negative consequences. This pattern suggests that those with the most to lose from faulty identification are already skeptical, while those who feel safer from both crime and misidentification are more supportive. The generational divide is particularly telling as younger Londoners will live with the consequences of normalized public surveillance far longer than their elders.
The Slippery Slope of Normalization
The planned expansion of LFR technology, including permanent installations in Croydon, represents a fundamental shift in policing methodology that deserves far more public debate. Once this infrastructure becomes normalized and widespread, the threshold for its use will inevitably lower. What begins as targeting serious offenders could easily expand to minor infractions, protest monitoring, or general public order maintenance. The department’s own report shows how quickly deployment scales – from zero to 203 deployments in a single year. Without robust legal safeguards, this technology could fundamentally reshape public space into a panopticon where citizens are constantly aware they’re being identified and tracked.
The Unaddressed Technical Limitations
Beyond racial bias, the report acknowledges other fundamental technical problems including poor lighting, bad angles, and obstructions causing misidentifications. The case of an identical twin being falsely identified reveals a deeper limitation: the technology cannot distinguish between genetically similar individuals, creating inherent risks for families. Similarly, the gender misidentification case shows the systems still struggle with basic classification tasks. These aren’t minor bugs that can be patched – they represent fundamental limitations in how computer vision interprets human features. When deployed in high-stakes policing contexts, these limitations become civil rights violations waiting to happen, particularly for communities already experiencing disproportionate police attention.
