{"resourceId":"biometric-jacksonville-frt-wrongful-20261001","versions":[{"version":"external-659e5eaf9d5d199a2600984022ee287ee92229222d33f763e3fadef811e07c3f","resource":{"id":"biometric-jacksonville-frt-wrongful-20261001","title":"Biometric Update: Jacksonville-area FRT wrongful arrests show probable-cause corroboration failures on FACESNXT","organization":"Biometric Update","sector":"Policing / facial recognition investigations","geography":"United States — Jacksonville / Jacksonville Beach, Florida (FACESNXT shared by 221 agencies)","publishedAt":"1 October 2026","publicationDate":"2026-10-01","eventDate":"2026-10-01","sourceName":"Biometric Update","sourceLabel":"Industry trade reporting on law-enforcement facial recognition misuse cases","sourceUrl":"https://www.biometricupdate.com/202610/with-wrongful-arrests-jacksonville-police-show-how-not-to-use-facial-recognition","evidenceClass":"independent-reporting","outcomeClass":"cautionary","topics":["governance-procurement","data-security","operating-model","knowledge-work"],"finding":"Joel R. McConvey’s 1 October 2026 article details Jacksonville Sheriff’s Office arrests based on FACESNXT facial-recognition candidates without adequate independent investigation—including a man jailed after an offender used his scraped Facebook photos in a predator sting, and Jalil Richardson’s arrest on an ~85% FRT match tied to a stolen-car investigation. Pinellas County Sheriff’s Office (FACESNXT operator) states facial recognition results are never “matches” and independent investigation is required; Florida law bars using an FRT match alone as probable cause. The piece frames repeated Jacksonville-area failures as a police process/guardrails problem more than a pure algorithm-accuracy story, noting FACESNXT’s ~15M mugshots/~25M DMV photos and 221 agency accessors.","sledRelevance":"Fresh caution for any SLED agency on shared face galleries: AI candidate lists without corroboration create wrongful-arrest and custody harms even when vendors/operators publish non-match rules.","evidence":"Opened Biometric Update HTTPS article via curl (HTTP 200) at 2026-10-04T02:50:30Z; confirmed FACESNXT scale figures, corroboration rule, and case narratives on page.","architectureImplications":"FRT outputs must be labeled candidates in RMS/case systems; block warrant auto-generation from score alone; require documented corroboration fields.","governanceImplications":"Enforce statutory non-PC rules with supervisor attestation before arrest/warrant; audit agencies on shared state systems like FACESNXT.","securityPrivacyImplications":"Scraped social photos and synthetic/stolen face risk means probe images need provenance checks; wrongful arrests impose lasting employment/custody harms.","caveats":"Trade publication summarizing WLRN/Florida Trib and WSOC reporting; defendant name withheld in one case; does not present a controlled accuracy study of Idemia algorithms.","streamIds":["public-safety"],"roles":{"sales":"Sales framing: shared face galleries fail customers when officers treat AI candidates as identification. Jacksonville’s repeated wrongful arrests—despite Pinellas stating results are never matches—show process controls, not just model specs, decide liability. Discovery should ask how warrants are blocked without corroboration and how probe-image provenance is checked when social-media photos may be scraped. Offer a bounded FRT investigative-guardrails assessment. Do not sell higher match scores as probable cause or claim FACESNXT-scale databases eliminate false attribution.","engineering":"Engineering framing: integrate candidate-only UI labels, mandatory corroboration checklist fields before warrant or arrest actions, and audit trails linking FRT scores to human investigative steps. On shared systems serving 221 agencies, home-agency policy must still gate local RMS actions even when the gallery operator publishes non-match rules. Proof of value is zero arrest-from-score-alone incidents in audit sampling, not marketing accuracy percentages such as an 85 percent candidate score.","delivery":"Delivery framing: rewrite FRT SOPs around independent investigation, train detectives on social-photo impersonation risks, and add IA sampling of FRT-initiated arrests. Ownership: investigations major with Pinellas/state gallery MOU constraints. Acceptance: attested corroboration on 100% of FRT-touched arrest packages in a pilot month; suppression of score-as-PC language in templates. Risks: cross-agency gallery access without local training; score anchoring; victim/offender photo spoofing."},"roleDetails":{"sales":{"customerProblem":"Agencies using shared FRT galleries risk wrongful arrests when candidates are treated as matches without corroboration.","stakeholders":["detectives","prosecutors","civil rights units","state gallery operators","defense bar"],"discoveryQuestions":["Can an officer seek a warrant on an FRT score alone in your RMS?","How do you verify probe-image provenance?","What audit samples FRT-initiated arrests?"],"valueHypothesis":"Enforcing candidate-only handling and corroboration gates reduces wrongful-arrest exposure without abandoning lawful FRT leads.","potentialEngagement":"FRT investigative controls review against statutory non-PC rules and shared-gallery MOUs.","unsupportedClaims":["That gallery size guarantees identity","That an 85% score is probable cause","That operator policy statements alone change field behavior"]},"engineering":{"fit":"Fits agencies querying FACESNXT-like shared face systems or vendor FRT plugged into case management.","architectureIntegration":"Candidate label + corroboration required fields before warrant/arrest status; immutable link of score, probe hash, and investigator notes.","prerequisites":["FRT use policy","RMS field customization","supervisor workflow","MOU with gallery operator"],"constraints":["multi-agency gallery access","social-media probe images","pressure for fast arrests in predator/stolen-auto cases"],"security":"Control who can run searches; log every query; restrict download of gallery images; warn on low-quality probes.","proofOfValue":["corroboration field completion rate","FRT-to-arrest audit with independent evidence listed","zero template language calling results matches"]},"delivery":{"work":"Update SOP, RMS validations, training on scraped-photo impersonation, and monthly FRT arrest audits.","dependencies":["prosecutor buy-in","gallery operator guidance","IA capacity"],"ownership":"Investigations command owns field compliance; records/IA owns audit evidence.","skillsAdoption":"Scenario training using Jacksonville-style failures; certify detectives before FRT query rights.","governanceCheckpoints":["policy re-issuance","first audit of FRT arrests","MOU review with Pinellas-like operator"],"acceptanceCriteria":["100% corroboration attestation on FRT-touched arrest filings in pilot","UI/templates say candidate not match","documented probe provenance checks"],"risks":["score anchoring","cross-jurisdiction training gaps","public trust loss after repeated wrongful arrests"]}},"retrievedAt":"2026-10-04T02:50:30Z","enrichedAt":"2026-10-04T02:50:30Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Training must reach every detective with gallery credentials, including smaller agencies among the 221 accessors.","procurementImplications":"Contracts/MOUs should require candidate labeling, audit logs, and local corroboration gates—not only algorithm accuracy claims.","operatingModelImplications":"Separate lead-generation from probable-cause formation; supervisors gate FRT-initiated liberty deprivations.","sourceVerification":{"openedUrl":"https://www.biometricupdate.com/202610/with-wrongful-arrests-jacksonville-police-show-how-not-to-use-facial-recognition","referenceExcerpt":"By law, a facial recognition match cannot be used as probable cause for an arrest without corroboration of evidence.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}