{"resourceId":"liberty-btp-lfr-zero-arrests-20260929","versions":[{"version":"external-581faa084529eb63f1e01693a9627c97f8a541b6fe4023a4b21e6ce7b0557975","resource":{"id":"liberty-btp-lfr-zero-arrests-20260929","title":"Liberty Investigates: BTP London rail LFR trial cost £320,786, scanned 500k+ faces, one false positive, no arrests","organization":"Liberty Investigates (with The Guardian)","sector":"Law enforcement / live facial recognition","geography":"United Kingdom — London railway stations (British Transport Police)","publishedAt":"29 September 2026","publicationDate":"2026-09-29","eventDate":null,"sourceName":"Liberty Investigates","sourceLabel":"FOI-based independent investigative journalism (also reported with The Guardian)","sourceUrl":"https://libertyinvestigates.org.uk/articles/live-facial-recognition-btp-trial-london-stations/","evidenceClass":"independent-reporting","outcomeClass":"cautionary","topics":["infrastructure","governance-procurement","data-security","operating-model"],"finding":"FOI material obtained by Liberty Investigates (published 29 September 2026) shows BTP’s six-month live facial-recognition trial (Feb–Jul 2026) across 18 London rail deployments cost £320,786 in equipment hire and staffing, scanned more than half a million faces, generated one watchlist alert that was a false positive, and produced no arrests from LFR alerts. BTP nonetheless extended the trial four months and expanded toward Underground stations with TfL VAWG framing. BTP says officers made associated arrests not caused by LFR alerts and is refining locations/watchlists. Former biometrics commissioner Fraser Sampson questioned proportionality given null catch yield; MP Bell Ribeiro-Addy called for suspension pending stronger law.","sledRelevance":"Newly relevant measured-null LFR evaluation (not previously archived) as U.S. and UK forces debate biometric expansion; transferability limited by UK legal context but the evaluation discipline is portable to U.S. SLED pilots.","evidence":"Opened Liberty Investigates HTTPS article via curl (HTTP 200); confirmed £320,786, 18 deployments, 500k+ faces, one false positive, zero LFR-alert arrests, extension, and BTP associated-arrest caveat. Evening Standard 2026-10-01 secondary confirmed continued same-week circulation.","architectureImplications":"Require outcome logging that separates alert-driven hits from nearby conventional arrests; gate site expansion on pre-registered yield thresholds.","governanceImplications":"Publish FOI-ready cost/alert/FP/arrest tables before extending LFR; do not let VAWG messaging substitute for alert performance evidence.","securityPrivacyImplications":"Mass face scanning on transit is special-category biometric processing; non-match retention and watchlist quality need explicit controls and oversight.","caveats":"UK policing/legal framework differs from U.S.; BTP disputes framing by citing associated non-alert arrests; Facewatch conflict note for Sampson’s current role.","streamIds":["public-safety"],"roles":{"sales":"Sales should open on measured alert yield and proportionality—not station coverage maps. Liberty Investigates’ FOI shows BTP’s six-month London rail LFR trial cost £320,786 across 18 deployments, scanned more than 500,000 faces, produced one watchlist alert that was a false positive, and zero arrests from LFR alerts—yet the trial was extended to Underground stations. Position any live-FR engagement as a bounded evaluation design with pre-registered success metrics and stop/go gates before expansion. Do not sell VAWG or wanted-offender outcomes from this pilot’s LFR-alert performance data.","engineering":"Engineering fit is time-boxed LFR deployments with watchlist governance, false-positive logging, and metrics that separate alert-driven arrests from nearby conventional policing. Integrate camera vans/kits with operator confirmation UI, retention limits for non-matches, and immutable alert outcomes. Prerequisites: legal basis, DPIA/oversight board review, controlled watchlist growth, and public evaluation criteria. Constraints: FOI-documented null yield, political VAWG framing that outruns alert performance, and ICO/oversight expectations. Proof of value: pre-declared minimum true-positive/arrest thresholds met before any extension.","delivery":"Delivery work is an evaluation governance package: publish FOI-ready cost/alert/arrest tables, require oversight sign-off before Underground expansion, and document associated arrests separately from LFR-alert performance. Ownership sits with force LFR lead and PCC/oversight board. Acceptance criteria should include a public metrics dashboard, false-positive case reviews, watchlist change logs, and a hard stop if alert-driven outcomes remain zero after a defined sample of deployments."},"roleDetails":{"sales":{"customerProblem":"Forces expand live facial recognition after pilots that show high scan volume and cost with near-zero alert-driven arrests.","stakeholders":["Chief constable / sheriff equivalent","PCC/oversight board","Data protection officer","Public transport partners","Civil society monitors"],"discoveryQuestions":["What pre-registered LFR success metrics justify extension?","How are false positives reviewed and disclosed?","Are associated arrests kept separate from alert performance?","What legal framework and oversight board hold deployments to account?"],"valueHypothesis":"Bounded value is evaluation-design and oversight tooling that prevents expansion on null alert yield—not claimed offender-catch rates from this FOI.","potentialEngagement":"4–8 week LFR evaluation redesign with FOI-ready metrics, watchlist controls, and go/no-go gates before further sites.","unsupportedClaims":["That £320k trial proved LFR effectiveness","VAWG reductions attributable to LFR alerts here","Zero privacy risk because non-matches are not retained like CCTV"]},"engineering":{"fit":"Operator-confirmed LFR with logged outcomes; unsuitable as always-on expansion without yield thresholds.","architectureIntegration":"Camera → watchlist match score → human confirmation → outcome log (TP/FP/no-alert) → retention policy for non-matches.","prerequisites":["DPIA","Watchlist data-quality rules","Oversight board","Public metrics schema"],"constraints":["Null alert-driven arrests in Feb–Jul window","Political expansion pressure","Special-category biometric processing"],"security":"Protect watchlists; minimize retention of non-match biometrics; secure FOI-responsive logging.","proofOfValue":["Alert-driven arrest threshold met","FP rate within policy","Independent oversight sign-off on extension"]},"delivery":{"work":"Stand up metrics reporting, FP case reviews, watchlist change control, and extension decision records.","dependencies":["Vendor kit contracts","TfL/station access","Legal/DPO review","Staffing for deployments"],"ownership":"Force LFR SRO owns operations; oversight board owns continuation decisions; DPO owns DPIA.","skillsAdoption":"Train operators that associated street arrests are not LFR performance; log alert causation explicitly.","governanceCheckpoints":["Mid-trial FOI metrics publish","Extension vote with yield data","Post-FP review"],"acceptanceCriteria":["Public cost/alert/arrest table","Zero undocumented watchlist adds","Stop rule if alert yield stays null"],"risks":["Mission creep to fixed cameras","Public backlash after FOI","Confusing associated arrests with LFR hits"]}},"retrievedAt":"2026-10-02T02:55:41Z","enrichedAt":"2026-10-02T02:55:41Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Operators must be trained to record whether an arrest was alert-caused; public messaging should not overstate LFR effectiveness to victims of VAWG.","procurementImplications":"LFR kit hire/staffing costs need evaluation milestones with stop rights when alert yield is null.","operatingModelImplications":"Pilot extensions should be formal go/no-go decisions tied to metrics, not automatic geographic expansion.","sourceVerification":{"openedUrl":"https://libertyinvestigates.org.uk/articles/live-facial-recognition-btp-trial-london-stations/","referenceExcerpt":"resulted in no arrests and just one alert, which turned out to be incorrect.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}