Public Safety & Emergency ServicesIssue 487 records
Public Safety
Tonight’s public-safety edition is about who controls AI and the data it runs on. Chile’s Supreme Court approved a policy allowing only commission-authorized AI uses, with human review and no AI deciding cases or weighing evidence. Covington, Ky., wrote an AI-training opt-out and retention of AI report drafts into its Axon agreement. On plate readers, Akron removed its Flock network from nationwide searches, Chicago police opened an audit of officers’ access to other agencies’ systems, and the ACLU warned about AI travel-pattern stops. Hampshire police reported 26 arrests from 406,001 faces scanned by live facial recognition, and a Brazilian court expanded facial recognition check-ins for alternative sentences.
- Evidence records
- 7
- Cross-source patterns
- 2
- Evidence classes
- 6 independent reporting1 government evaluation
- Outcomes
- 3 emerging3 cautionary1 mixed
- Source freshness
- 7 new this fortnight
- Research completed
- 2026-10-27
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Plate-reader governance is moving from whether to keep cameras to who may query the data and what a query or alert can justify
Akron kept its Flock cameras but cut out nationwide searches after its logs showed most outside queries were blanket searches. Chicago, which relies on other agencies’ networks, cut access to job-required roles and proposed a documented purpose for every query plus monthly audits. The ACLU argues the next problem is AI pattern analysis that flags drivers as suspicious and leads to pretext stops. All three treat the query and its basis as the unit of control. None provides effectiveness data, Akron and Chicago published no audit counts, and the ACLU piece is advocacy built on anecdotes.
Operating questionFor every plate-reader search or AI alert last month, could you show who ran it, why, and whether it led to a stop?
Supporting evidenceCity of Akron / Akron Police DepartmentChicago Police DepartmentAmerican Civil Liberties Union (Speech, Privacy, and Technology Project)
AI controls in justice are being written into institutional authorizations and contract terms rather than left to individual users
Chile’s Supreme Court approved a policy under which only uses authorized by a court commission are allowed, with human review required and AI barred from deciding cases or weighing evidence. Covington, Ky., wrote an AI-training opt-out and retention of AI-drafted report narratives into its Axon master agreement. Both shift control from individual judges or officers to an institution-level rule that can be audited. They differ in force: Chile’s text was not yet published and is reported from court sources, while Covington’s terms are described by the city without the contract itself.
Operating questionIs each AI use in your agency covered by a written approval or contract clause, or does it depend on individual users’ judgement?
Supporting evidenceCorte Suprema de Chile (Poder Judicial de Chile)City of Covington, Kentucky / Covington Police Department
Evidence ledger
Full record · every source keeps its link and limitations
Chile’s Supreme Court approves its first institutional AI policy: only authorized uses, mandatory human review, and no AI deciding cases or weighing evidence
La Tercera reports that the full Chilean Supreme Court, closing its annual reflection sessions on Friday (Oct. 23), approved the judiciary’s first institutional AI policy, drafted by its Studies Directorate and Justice Jean Pierre Matus; the text is to be published in the coming days. According to sources who have seen it, the policy moves from individual use to institutional use: in judicial work and court systems only AI uses authorized by the Electronic Processing Commission (Comisión de Tramitación Electrónica) will be allowed, each authorization specifying purpose, data and conditions. Entering draft rulings or personal data into unauthorized AI tools will be prohibited. AI may never decide a case, replace the weighing of evidence or write the reasoning of rulings; every authorized use requires human review. The policy adds a risk-grading table, a catalogue of authorized systems and uses, and an AI Committee with a Supreme Court justice and staff association representatives. A companion La Tercera article the same day quotes Court President Gloria Ana Chevesich saying the instrument has 13 guiding principles, including human oversight, explainability, traceability and judicial independence.
Why it matters, evidence and limitations
- Why it matters
- A national court system replacing individual judge-by-judge AI use with an approved-use catalogue, a risk table and a standing committee, a governance model U.S. state court administrators are also weighing.
- Evidence and measured results
- Opened the La Tercera article and read it in full; cross-checked against La Tercera’s companion article quoting the Court President’s Monday briefing.
- Limitations and uncertainty
- The policy text was not yet published when retrieved; details come from press reporting based on court sources and the President’s briefing. No data on current AI use or error rates. Chilean civil-law context limits direct transfer to U.S. courts.
Akron removes its Flock plate-reader network from nationwide searches after finding most outside searches were blanket queries
Akron announced on Oct. 26, 2026 a policy, effective immediately, that removes its Flock camera network from nationwide searches by outside agencies. Akron police and agencies within Ohio keep access; out-of-state agencies with a legitimate investigative need must request access or data directly from Akron police. The city said its ongoing review of Flock system logs found the "vast majority" of outside searches were nationwide blanket searches, not targeted requests. Mayor Shammas Malik said those searches provide little to no public safety benefit to residents but create openings for uses that conflict with city policy. The city had already blocked immigration-enforcement and reproductive-health searches in May. Council members have questioned camera placement and one proposed suspending the program; the public safety chief cited three cases where cameras helped. The Flock contract runs through 2028.
Why it matters, evidence and limitations
- Why it matters
- A city using its own audit logs to narrow data sharing rather than cancel its ALPR program, a middle path between full Flock exits and the status quo.
- Evidence and measured results
- Opened the Ideastream article and read it in full.
- Limitations and uncertainty
- Based on a city news release as reported; the log review figures behind "vast majority" were not published in the source. No data on whether opting out affects case outcomes. Earlier Akron coverage (Sept. audit, Oct. 19 council debate) is archived separately.
Chicago police audit officers’ use of other agencies’ plate-reader systems, cut access and propose documented-purpose and monthly-audit rules
The Chicago Police Department is auditing whether officers abused access to other law enforcement agencies’ license plate reader systems. It has immediately limited access to outside systems to members whose position and job duties require it, and each of its five bureaus must designate someone to oversee ALPR use, including outside systems. Proposed changes in a draft directive open for public comment through Nov. 12 include non-transferable logins, a documented legitimate law enforcement purpose for every query, monthly search audits, central processing of access requests by the Office of Public Safety Administration, and penalties from loss of access to referral for prosecution. The department did not say what triggered the audit. The article notes felony charges against officers in several Illinois and Wisconsin agencies for Flock misuse and a Will County Board vote on an ALPR ban set for Nov. 12.
Why it matters, evidence and limitations
- Why it matters
- A large department that does not own Flock still relies on other agencies’ networks, so its controls must cover borrowed access, not just its own cameras.
- Evidence and measured results
- Opened the Sun-Times article and read it in full.
- Limitations and uncertainty
- The department did not explain what prompted the audit or when it will finish; no count of suspected misuse is given. Proposed rules may change after comment. Earlier coverage of CPD’s outside Flock searches (Oct. 7) is archived separately.
Covington, Ky., signs a 10-year Axon agreement that opts out of AI training on city data and requires retention of AI-drafted police report narratives
Covington announced a ten-year Master Services and Purchasing Agreement with Axon for its police department, with annual renewals subject to appropriation. It covers Fleet 3 in-car cameras (partly paid with civil asset forfeiture funds, with no ALPR purchased), body-worn cameras, TASERs, virtual reality training and Axon Air drone-as-first-responder equipment, with scheduled replacements. The city says it expressly opted out of Axon’s voluntary data-sharing, customer experience, product enhancement and AI training programs, which bars Axon from using city data, including anonymized or aggregated data, to develop products or train shared AI models, with limited exceptions. For AI-generated police reports, Axon must retain and make available the original AI-generated draft narratives and the final officer-edited reports, and may not auto-delete them before the city’s retention period ends unless the city directs it in writing.
Why it matters, evidence and limitations
- Why it matters
- A city writing AI-specific protections into a vendor master agreement: no training on its data and preservation of AI drafts so edits can be compared, addressing a criticism that some report tools discard drafts.
- Evidence and measured results
- Opened the city news release and read it in full.
- Limitations and uncertainty
- City release only; the contract value, AI report tool name and pilot results are not stated. No evidence about report accuracy or time savings. The exceptions to the opt-out are described only briefly.
ACLU warns AI analysis of plate-reader data is generating "suspicious travel pattern" stops of drivers with no prior suspicion
In an Oct. 26, 2026 commentary, Jay Stanley of the ACLU Speech, Privacy, and Technology Project argues that AI analysis of ALPR data is being used to flag drivers’ travel patterns as suspicious and to justify pretextual stops. He recounts a previously unpublished account of a driver stopped in 2025 by an officer who said plate scans showed suspicious travel patterns and showed her a map of her plate hits. He cites Flock’s 2025 marketing of pattern and convoy analysis, reporting on Flock’s next-generation AI querying plate data with other records, AP reporting on Border Patrol travel-pattern monitoring and a Texas jury verdict for a driver stopped after a fusion center alert. He argues this shifts ALPR use from investigating suspicious vehicles to generating suspicion, and calls for ending license plate surveillance.
Why it matters, evidence and limitations
- Why it matters
- Identifies the next governance question for ALPR programs: not just who can search, but whether AI pattern alerts may be the basis for a stop.
- Evidence and measured results
- Opened the ACLU commentary and read it in full.
- Limitations and uncertainty
- Advocacy commentary; the central account is one person’s recollection and the officer’s identity and system are unverified. Other cases are cited from secondary reporting. No data on how often AI pattern alerts lead to stops.
Hampshire police live facial recognition: 406,001 faces scanned over 20 deployments, 69 confirmed alerts and 26 arrests
Using Hampshire Constabulary data, the Daily Echo reports that live facial recognition scanned 406,001 faces in 20 deployments across eight towns between December 2025 and October 2026, producing 69 confirmed alerts and 26 arrests, about one arrest per 15,600 faces. Four Southampton deployments scanned 61,394 faces with 20 alerts and nine arrests. The force says no false alerts have been created in Hampshire, non-matching images are deleted immediately and it will keep using the technology; the police and crime commissioner notes the vans are Home Office funded. Liberty and Big Brother Watch say the yield does not justify the scale, noting there is no dedicated law and citing a local wrongful arrest from a retrospective facial recognition search.
Why it matters, evidence and limitations
- Why it matters
- Rare force-level yield figures for live facial recognition that both sides read differently: arrests per deployment versus faces scanned per arrest.
- Evidence and measured results
- Opened the Daily Echo article and read it in full; the force’s results page returned HTTP 403.
- Limitations and uncertainty
- Secondary report of force data; the force’s own page was not accessible. "No false alerts" is the force’s claim and depends on its definition of confirmed alert. No data on charges, convictions or watchlist composition. UK context.
Tocantins court expands AI facial recognition remote check-ins for people serving alternative sentences to a third district
The Tocantins Court of Justice launched SAREF (remote check-in with facial recognition) in Gurupi on Oct. 26, 2026, its third district after Palmas and Araguaína. People serving alternative sentences who must report periodically can check in by phone with facial recognition and geolocation, or at a kiosk in the courthouse, after in-person enrolment with personal data and a photograph. Facial recognition confirms identity and the system records the check-in for later validation by the court unit. SAREF was developed by the Federal District court (TJDFT) and adapted for national use on the national judiciary digital platform (PDPJ-Br) under the CNJ Justiça 4.0 programme. The court presents it as reducing travel without weakening supervision.
Why it matters, evidence and limitations
- Why it matters
- A court-run biometric check-in for community supervision, comparable to U.S. probation remote reporting apps, with human validation of each check-in.
- Evidence and measured results
- Opened the Portal Agora Tocantins article and read it in full.
- Limitations and uncertainty
- Regional news based on a court announcement; no data on usage, match errors, missed check-ins or compliance outcomes. Brazilian legal context; transfer to U.S. probation requires attention to biometric privacy laws.
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.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Government evaluation
- A public body’s measured evaluation or documented pilot.