Metropolitan Police LFR annual report: scans, alerts, arrests and denominators

Primary source: Metropolitan Police, Live Facial Recognition Annual Report, 30 September 2025, covering 11 September 2024–10 September 2025. See state-power pathway. This is a police operational self-report, not an independent measurement of crime prevented. A later Met Croydon pilot release (13 May 2026) reports 24 static-camera operations in October 2025–March 2026, 470,000+ passers-by, 173 arrests and one false alert; do not merge its denominators with this annual period.

What it counts

207 deployments (203 crime hotspots, four event-protection), estimated 3,147,436 faces passed cameras, 2,077 alerts including 2,067 judged true and 10 false; six of the false alerts led to officer engagement, none to arrest. The observed 10/3.147m is ~0.0003% per face seen, not the chance that a watchlisted person is caught, nor the probability that someone stopped is a suspect. The Met reports 962 arrests: 549 linked to court warrants, 347 to police suspicion, 85 to order/restriction breaches (the subgroups overlap; some people arrested more than once). Of the 347 police-suspected arrests, 146 charged, 88 bailed pending further investigation, 113 no further action. NFA does not prove an unlawful arrest, but arrest count is not conviction count. The report also records 1,084 encounters with monitored/order-constrained persons, including 1,050 registered sex offenders; 213 were in breach but not arrested. Their presence on a watchlist did not itself mean arrest was warranted.

Typical hotspot watchlist about 10,250 people, with courts, suspected serious crime, MAPPA and court orders represented; categories overlap. Watchlists assembled at most 24 hours before deployments and deleted afterwards per Met report. A claimed 10.5% Croydon crime drop versus previous year is not causal evidence without a suitable control; arrestees include people wanted for serious sexual offences and long-standing warrants. In June 2026 Met announced plans to expand static camera coverage, which changes the future deployment surface.

Limits worth retaining

Counterfactual arrests without LFR and crimes averted are unmeasured; false negatives (watchlisted people passing unnoticed) are not observable from its log, and the probability of unnecessary police contact depends on alert adjudication. Watchlist eligibility is a political/organizational decision separate from precision. Met self-report should be checked against the NPL and ethics panel; the ICO had not yet audited the Met as of its August 2026 five-force overview.