{"resourceId":"wa-police-live-facial-recognition-trial-1-38m-faces-122-arrests-9-false-alerts-20261102","versions":[{"version":"external-f35fed114c00014881201023b371f5acf19d7e1eb4c194c2b451f935d9dd9d60","resource":{"id":"wa-police-live-facial-recognition-trial-1-38m-faces-122-arrests-9-false-alerts-20261102","title":"Western Australia Police end a five-month live facial recognition van trial: 1.38 million faces scanned, 122 arrests, nine false alerts, and a disputed error rate","organization":"Western Australia Police Force","sector":"Policing / live facial recognition","geography":"Western Australia, Australia","publishedAt":"The Guardian (Australia), first published Nov. 1, 2026 (U.S. Eastern), updated Nov. 2, 2026","publicationDate":"2026-11-01","eventDate":"2026-10-31","sourceName":"The Guardian","sourceLabel":"Police use facial recognition to scan one million faces in WA trial leading to 122 arrests","sourceUrl":"https://www.theguardian.com/australia-news/2026/nov/02/wa-police-facial-recognition-technology-trial","evidenceClass":"independent-reporting","outcomeClass":"mixed","topics":["data-security","governance-procurement","operating-model"],"finding":"The Guardian reported that Western Australia Police ended a five-month live facial recognition trial on Saturday, Oct. 31, 2026. Police said vans and vehicle-mounted cameras scanned 1,380,882 faces across 100 deployments at 43 locations, matching against an alert list from police databases. Police reported 122 arrests, engagement with 146 registered sex offenders, 10 welfare checks, and nine incorrect alerts, which they described as an error rate of 0.00065% of scans. The commissioner said no one wrongly alerted was arrested but each was approached and questioned. Civil Liberties Australia argued the rate should be judged by how many arrests lead to convictions. The state information commissioner said it was not involved in the trial. Vans are withdrawn while police run an internal evaluation of performance, community impact and safeguards.","sledRelevance":"A large, fully reported live facial recognition trial with operator-supplied outcome counts, showing which metrics agencies publish and which they leave out.","evidence":"Opened and read the full Guardian article. All counts are police figures as reported; the internal evaluation and any underlying data were not published.","architectureImplications":"Live matching against a watchlist with real-time pixelation and deletion of non-matches; the watchlist composition drives both hits and false alerts.","governanceImplications":"The trial ran without the privacy regulator’s involvement, and the evaluation is internal. The denominator used for the error rate is contested.","securityPrivacyImplications":"Police say non-matching images were deleted immediately; this is not independently verified. Every false alert led to a stop and questioning.","caveats":"Australian law and policing differ from U.S. practice; figures are police self-reports; arrest outcomes and demographic error data were not reported; the vendor is not named in the article.","streamIds":["public-safety"],"roles":{"sales":"Sales framing: Western Australia gives agencies a concrete set of live facial recognition numbers, but they are police figures and the key accuracy measure is disputed. The buyer problem is proving a trial worked to a skeptical public. A bounded engagement could design the evaluation metrics before deployment: precision per alert, stop outcomes, and court results. Do not cite the 0.00065% figure as accuracy; it divides false alerts by all faces scanned, not by alerts.","engineering":"Engineering framing: the system matches live camera feeds against a police watchlist and pixelates and deletes non-matches in real time. Pre-sales should ask how the watchlist is built and reviewed, how deletion is verified, and whether every alert, stop and outcome is logged. A proof of value would measure precision per alert and demographic false-match rates on local data. Constraints: watchlist quality, lighting and crowd conditions, and regulator involvement.","delivery":"Delivery framing: the work is running a time-limited trial with an evaluation designed in advance and published afterwards. Ownership sits with a senior police lead, with privacy counsel and an external reviewer. Acceptance criteria are proposals: published precision per alert, outcomes for every person stopped, deletion audits, and a go or no-go decision before redeployment. Main risks: an internal-only evaluation, watchlist drift, and public backlash over stops of people wrongly alerted."},"roleDetails":{"sales":{"customerProblem":"Agencies need to show whether live facial recognition is accurate and proportionate, but trial reporting often uses favourable metrics.","stakeholders":["Police chief or commissioner","Privacy officer or regulator liaison","Legal counsel","Community oversight body"],"discoveryQuestions":["What share of alerts were correct, not just what share of scans?","Who builds and reviews the watchlist?","Will the evaluation be published and independently reviewed?"],"valueHypothesis":"A pre-registered evaluation lets leaders decide on expansion with defensible numbers.","potentialEngagement":"Evaluation design and independent review for a time-limited facial recognition trial.","unsupportedClaims":["Accuracy based on false alerts divided by all scans","Crime reduction from the trial","Absence of bias, which was not reported"]},"engineering":{"fit":"Relevant to agencies weighing live or mobile facial recognition.","architectureIntegration":"Camera feeds to a matching service against a police watchlist, with real-time deletion of non-matches and alert dispatch to officers.","prerequisites":["Documented watchlist criteria","Alert and stop logging","Deletion verification"],"constraints":["State or national biometric law","Regulator consultation","Crowd and lighting conditions"],"security":"Verify deletion of non-matches, restrict watchlist access, and log every alert with the officer response.","proofOfValue":["Precision per alert on local deployments","Demographic false-match test","Deletion audit sample"]},"delivery":{"work":"Design the evaluation, run the trial, publish results, and decide on redeployment.","dependencies":["Watchlist governance","Records of stop outcomes","Court outcome data"],"ownership":"Senior police lead with privacy counsel and an external reviewer.","skillsAdoption":"Officers need clear procedures for approaching people after an alert and recording the result.","governanceCheckpoints":["Evaluation plan approved before launch","Regulator consulted","Published results before expansion"],"acceptanceCriteria":["Precision per alert reported","Every stop outcome recorded","Results published within a set period after the trial"],"risks":["Internal-only evaluation","Watchlist drift","Public loss of trust after wrongful stops"]}},"retrievedAt":"2026-11-03T03:54:34Z","enrichedAt":"2026-11-03T03:57:16Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Officers must follow up every alert in person, so alert volume becomes a staffing load.","procurementImplications":"Contracts should require watchlist governance, measured precision per alert, and outcome tracking through to court, not only scan counts.","operatingModelImplications":"Pausing deployment for evaluation before expansion is a sound gate; its value depends on whether the evaluation is published.","sourceVerification":{"openedUrl":"https://www.theguardian.com/australia-news/2026/nov/02/wa-police-facial-recognition-technology-trial","referenceExcerpt":"the system generated nine incorrect alerts against the total 1.38m scans, meaning there was an error rate of 0.00065%.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}