No independent pre-deployment testing infrastructure for police surveillance algorithms

The only substantive UK accuracy and bias evaluation of police live facial recognition is the 2023 National Physical Laboratory study commissioned by the Met and South Wales Police themselves, and subsequent Home Office equitability reports were published buried inside the December 2025 facial recognition plan. No UK equivalent of the US NIST's continuous face-recognition vendor testing exists, no requirement makes algorithms pass independent evaluation before operational use, and the ICO's 2025 police audit was retrospective and governance-focused. Police meanwhile run rapidly growing volumes of retrospective searches against passport and immigration databases, per Privacy International in 2025, with no published accuracy figures or audit.

Deployment decisions rest on vendor claims and one police-commissioned study. Without standing independent evaluation, demographic error differentials and the drift into passport-database searches go unmeasured, leaving public debate and litigation evidence-poor at the moment of national rollout.

A statutory pre-deployment testing regime: biometric and algorithmic policing tools must pass independent evaluation at a designated facility, the NPL or the Alan Turing Institute, with published per-deployment performance and demographic-differential statistics, plus a standing audit of searches against passport and immigration databases. A pilot facility at £5-10m is fundable ahead of the expected biometrics bill.

// Build together: Counterparty: host lab (NPL/Turing) plus vendors and forces submitting algorithms; pilot facility fundable now, statutory regime is the end-state.

Deployment decisions rest on vendor claims and a single police-commissioned study, and the independent testing facility is buildable, supporting in nature, and undated.