Public Safety & Emergency ServicesIssue 615 records
Public Safety
This public-safety edition tracks ALPR vendor exits and facial-recognition rollouts. Fresno will not renew Flock while keeping city-owned readers and writing an ordinance. University at Buffalo paused an LPR pilot for campus input. A California appellate court allowed ChatGPT-assisted expert testimony used for organization. Avon and Somerset set a Bath live facial-recognition debut, and Hong Kong enabled AI FR on SmartView CCTV with long metadata retention.
- Evidence records
- 5
- Cross-source patterns
- 2
- Evidence classes
- 3 government evaluation2 independent reporting
- Outcomes
- 3 emerging2 cautionary
- Source freshness
- 5 new this fortnight
- Research completed
- 2026-11-07
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Agencies are keeping ALPR capability while changing vendors or pausing for governance, not ending plate reading
Fresno will not renew Flock’s service yet will seek another company to run city-owned readers and write a citywide ordinance after misuse and out-of-state access concerns. University at Buffalo indefinitely paused its LPR pilot so an advisory committee can set use and data-handling rules before any install. In both cases the contested object is the vendor relationship and public process, while investigative plate-reading capacity remains on the table.
Operating questionIf we change ALPR vendors or pause a pilot, what ordinance, audit and data-handling rules will be in place before the next camera goes live?
Supporting evidenceCity of Fresno / Fresno Police DepartmentUniversity at Buffalo
Facial recognition deployments are expanding with human authorisation claims but thin published accuracy metrics
Avon and Somerset announced overt Live Facial Recognition with officer review, deletion of non-matches and PCC scrutiny, yet published no false-positive or demographic performance figures before the Bath launch. Hong Kong Police will enable AI facial recognition on SmartView CCTV with claimed case-triggered activation and inspector-level controls while retaining facial metadata up to 180 days and linking mall feeds, citing pilot anecdotes rather than independent evaluations. Authorisation language is outrunning measured evidence.
Operating questionBefore any FR go-live, which accuracy, false-positive and demographic metrics will we publish, and who can stop a deployment that misses them?
Supporting evidenceAvon and Somerset Police / Office of the Police and Crime CommissionerHong Kong Police Force
Evidence ledger
Full record · every source keeps its link and limitations
Fresno will not renew Flock Safety ALPR contract after misuse and sharing concerns; city will seek another vendor and a citywide ordinance
On Nov. 6, 2026 Fresno Mayor Jerry Dyer announced the city will not renew its Flock Safety ALPR service contract when it expires in February, citing ineffective safeguards despite vendor modifications, while keeping city-owned cameras and seeking another vendor with stronger safeguards for likely December council approval. Police Chief Mindy Casto said more than 70 fixed readers helped solve 42 shootings and 16 murders from January–October 2026 (agency-claimed, uncontrolled). A review found out-of-state agencies had accessed Fresno Flock data despite sharing settings, including “immigration” as a search reason, though officials said they found no ICE searches and could not determine whether access came from a department setting or a Flock change. An officer is under investigation for alleged non-investigative abuse. Councilmember Nick Richardson plans a citywide ordinance with search-record, 30-day retention and oversight rules. The three-year Flock contract cost about $1.5 million via a state Organized Retail Theft Prevention Grant. Flock said audit assistance, case-code requirements by end-2026, and proactive lockouts are coming.
Why it matters, evidence and limitations
- Why it matters
- A major California city is ending a Flock service relationship over misuse and cross-agency access doubts while still buying ALPR capability—useful for SLED buyers comparing vendor lock-in, audit controls and ordinance design.
- Evidence and measured results
- Opened the Fresno Bee article via WebFetch after curl to the origin failed/timed out; saved full article text. Corroborated with the same-day KMPH FOX26 local report opened via HTTPS curl (User-Agent SLED-research-bot/1.0) and converted to text. Key figures and quotes matched across both accounts.
- Limitations and uncertainty
- Crime-reduction attribution is agency-claimed and uncontrolled; Fresno Bee page blocked direct curl so text came via WebFetch; replacement vendor unnamed; officer investigation details withheld.
California Fourth District Court of Appeal finds no abuse of discretion admitting forensic psychologist testimony prepared with ChatGPT used as an organizational “thought partner”
On Nov. 5, 2026 the California Courts Newsroom reported that Division One of the Fourth District Court of Appeal held a trial judge did not abuse her discretion by allowing a forensic psychologist’s testimony after she admitted using ChatGPT as a “thought partner” to prepare a report opposing pretrial mental health diversion, where the judge credited that the tool was used for organizational purposes. The underlying unpublished opinion in Gillespie v. Superior Court (D088222), filed Nov. 4, 2026, denied writ relief. The defense argued AI assistance required Kelly/Frye proof and undermined independence; the appellate court deferred to the trial court’s credibility finding that AI did not impair the independence of the expert opinion, while agreeing in principle that AI use by a forensic expert can raise novel-technique concerns. The expert testified she used ChatGPT under a setting that excluded her material from the training model, for organization/polish, and denied using it to form the nexus opinion. The opinion is unpublished and non-citable under California Rules of Court rule 8.1115(a).
Why it matters, evidence and limitations
- Why it matters
- U.S. trial and appellate courts are starting to police AI use in expert reports through credibility and disclosure, not categorical exclusion—relevant to prosecutors, defenders and court IT policies on generative AI in casework.
- Evidence and measured results
- Opened the California Courts Newsroom item via HTTPS curl and converted to text. Also downloaded and read the full unpublished PDF opinion D088222 from courts.ca.gov (pdftotext) for holdings on ChatGPT use, Kelly/Frye arguments and the trial court’s credibility finding.
- Limitations and uncertainty
- Opinion is unpublished and non-citable; fact-specific credibility finding; does not create a general California rule authorizing ChatGPT in expert work; diversion denial also rested on substantial-evidence grounds beyond AI.
Avon and Somerset Police announce Live Facial Recognition with first Bath deployment planned for 25 November, with South Wales support and PCC scrutiny
On Nov. 5, 2026 Avon and Somerset Police said it will introduce Live Facial Recognition in coming weeks, with the first deployment planned in Bath on Wednesday 25 November subject to final operational authorisation. South Wales Police will support an initial deployment using its equipment and expertise. LFR compares live CCTV faces on a closed network to a pre-authorised watchlist; officers review alerts before action; non-matches are automatically and permanently deleted; watchlists are deleted after each deployment. Deployments are to be overt with marked vehicles, signage outside the Zone of Recognition and officers available to the public. Senior officer authorisation and equality/data-protection assessments are required for each use. The PCC’s office led community, academic and resident engagement to shape scrutiny. No measured hit rates, false-positive rates or demographic performance figures are published in the announcement.
Why it matters, evidence and limitations
- Why it matters
- A transferable UK force-level LFR rollout with human review, deletion defaults and PCC scrutiny—useful contrast for U.S. SLED agencies debating watchlist FR under civil-rights constraints.
- Evidence and measured results
- Opened the official Avon and Somerset Police news page via HTTPS curl (User-Agent SLED-research-bot/1.0) and converted HTML to text; read in full.
- Limitations and uncertainty
- UK legal context differs from U.S.; announcement is prospective (subject to authorisation); no published accuracy or bias metrics; vendor stack not fully specified.
University at Buffalo indefinitely pauses planned campus Flock/LPR pilot so Police Advisory Committee can gather community input
On Nov. 5, 2026 University at Buffalo published that President Caroline Attardo Genco directed an indefinite pause of plans for a license plate reader pilot for University Police, to allow more campus conversation. The University Police Advisory Committee (faculty, staff, students) will engage the community and then advise senior leadership on appropriate use, data security/handling, and whether to proceed. Local reporting (Spectrum News, Nov. 6) described the paused plan as a ~60-day pilot of about 20 Flock cameras on North and South campuses; the official UBNow brief does not itself name Flock or camera counts. No outcome metrics exist because the pilot did not start.
Why it matters, evidence and limitations
- Why it matters
- A public university halted an ALPR pilot for governance and data-handling review before install—useful for campus and municipal buyers under pressure to skip public process.
- Evidence and measured results
- Opened the official UBNow brief via HTTPS curl and converted to text. Corroborating Spectrum News local article also opened; vendor/camera-count details treated as secondary reporting not stated in the official brief.
- Limitations and uncertainty
- Official brief does not name Flock or give camera counts; those details come from secondary local reporting; pause is indefinite with no decision date; no technical evaluation yet.
Hong Kong Police to enable AI facial recognition on SmartView CCTV this month for criminal searches and high-risk missing persons; FR metadata retained up to 180 days
On Nov. 6, 2026 RTHK reported Hong Kong Police will roll out facial recognition across its CCTV network this month to assist criminal searches and locate high-risk missing persons. About 10,000 SmartView police cameras plus ~10,000 feeds from other government/public organisations are in place; FR can be activated on police and integrated third-party feeds but is said to be triggered for active searches rather than continuous 24-hour analysis on all cameras. Pilot anecdotes include a robbery arrest and locating an elderly man with dementia within 1.5 hours. CCTV footage retained max 31 days; FR photos and metadata (location, date/time, facial characteristics) retained 180 days. At least four shopping malls agreed to connect feeds; DOJ advice reportedly makes disclosure of police connection voluntary though signage that recording is underway is required. Target ~60,000 cameras by 2028. Force claims nearly 2,200 cases solved and >2,800 arrests since CCTV project began (uncontrolled before/after). HKFP same-day reporting adds that activation requires inspector-or-above sign-off and that police say mall links comply with privacy assessments.
Why it matters, evidence and limitations
- Why it matters
- Large-scale FR-on-CCTV with inspector approval and long FR-metadata retention is a cautionary international comparator for U.S. SLED debates on live FR, mall partnerships and retention clocks.
- Evidence and measured results
- Opened the RTHK English article via HTTPS curl and converted to text. Corroborated with same-day Hong Kong Free Press reporting opened via curl for inspector-approval and privacy-assessment claims.
- Limitations and uncertainty
- Hong Kong legal/political context differs sharply from U.S.; case-solved and pilot anecdotes are police-claimed; independent accuracy/bias evaluation not opened; mall list from RTHK understanding.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.