Here is a state-power harm that requires no rogue AI: police choose the watchlis
Here is a state-power harm that requires no rogue AI: police choose the watchlis
The Met reports 207 operations, 2,067 true alerts and 962 arrests during 11 September 2024–10 September 2025. Among the arrests were 549 linked to court warrants and 347 to police suspicion (categories overlap); 113 of the latter led to no further action, which does not itself establish an improper arrest. A companion May 2026 police report describes fixed-camera deployment in Croydon; camera access need not mean a citywide always-on system.
Officers adjudicate matches, unmatched biometric templates are deleted under published policy, and camera deployments are time and place bounded—important limits on erroneous or ubiquitous tracking. Yet an independent London Policing Ethics Panel review observed watchlist growth and broad eligibility, and called for stronger evidence of net public benefit and repeat-deployment effects. Arrest totals and the Met’s year-on-year local crime change cannot isolate what LFR prevented versus what ordinary policing would have achieved. Neither a false-alert denominator nor a poll on stated event avoidance measures the full rights trade-off. Next test: independent audit of who went on each list, officer stops, counterfactual crime and exposure at repeated sites.