Legal basis

Varies by state; no dedicated federal ALPR statute

Oversight

Fragmented. State-level audits in some cases (e.g. Illinois), but no single national body.

Fragmented by design: thousands of individual agencies each contract with Flock separately, with far less national standardisation of retention or access than a single national system would have. Scale as of August 2026 is roughly 120,000 cameras across 49 states, with a company-cited reach of 6,000 to 7,000 law enforcement agencies; sources differ on the exact headcount and the figures move quickly. That fragmentation is where the most serious documented problems have surfaced. An Illinois state audit found federal immigration authorities had gained access to state plate-read data through an undisclosed CBP arrangement, in apparent tension with state law, and 47 agencies were subsequently removed. A Dayton, Ohio audit in May 2026 found camera data searched more than 7,100 times for immigration enforcement in breach of city policy; the city covered all 72 cameras with bin bags pending contract exit. In San Francisco, 299 improper queries were made over about a year by a regional intelligence centre. The University of Washington Center for Human Rights documented federal access to Washington state networks through front-door, back-door and side-door routes. The Electronic Frontier Foundation reported over eighty agencies entering search justifications using language perpetuating stereotypes about Romani people. Individual officer misuse is now documented repeatedly: a Washington Post investigation found dozens of officers using the network to surveil former partners and romantic interests, and separate 2026 cases include a Georgia officer arrested in August, an Arizona officer who resigned after allegedly tracking his wife, and a North Carolina officer charged over 31 searches of her boyfriend's ex-wife's vehicle. Vendor response, 13 August 2026: default retention cut from 30 days to 7; the previously optional abnormal-search monitoring tool made universal, having been enabled by only about a third of customers; automatic lockouts for suspicious search patterns; case numbers required for extended retention. Flock had earlier removed federal agencies from its national search network (August 2025) and required a NIBRS offence type on every search (December 2025). Contraction is substantial but so is growth: around 53 cities had cancelled by late June 2026 and over 20 jurisdictions moved to drop Flock in July 2026 alone, while roughly 800 US cities approved new contracts during 2026. Ring cancelled a planned Flock integration in March 2026. Washington State's Driver Privacy Act (ESSB 6002), signed 30 March 2026 and effective immediately, imposes a 21-day retention cap, permitted-purpose limits, bans on immigration and reproductive-healthcare and protected-activity uses, collection bans near sensitive sites, mandatory registration with the Attorney General and annual audits.

Source: Company statements; Illinois state audit findings, reported by State of Surveillance and the ACLU · last checked 2026-08-25

Legal basis

City council authorisation

Oversight

City council vote following public disclosure

Police Chief Mike Canfield disclosed to the city council in February 2026 that federal agencies had gained access to the city’s Flock camera data without the city’s knowledge. The council voted unanimously to terminate the contract and the department discontinued use. Included as a specific, well-documented instance of a US agency withdrawing a deployment after an oversight failure, distinct from the wider Flock Safety entry tracked separately.

Source: City council disclosure, reported by State of Surveillance and the ACLU · last checked 2026-08-25

Legal basis

Varies by state; several states and 20+ cities have banned or restricted police facial recognition

Oversight

Fragmented. State and city-level restrictions in some jurisdictions, but no single federal framework.

Clearview markets its database of images scraped from the web as an investigative lead generation tool, and states that results are indicative rather than definitive matches requiring further verification. The roughly 3,100 US agency figure originates from the company itself and should be read as a historical vendor estimate, not a verified current count. A caution on attribution that matters for accuracy: not every US facial recognition wrongful arrest is a Clearview case, and this entry should not absorb them. The documented Clearview-attributed cases are Randal Quran Reid (Georgia, arrested for a Louisiana offence he had no connection to) and a Jefferson Parish Sheriff's Office case that led to a $200,000 settlement. By contrast, the June 2026 Florida suit filed by the ACLU on behalf of Robert Dillon concerns Pinellas County's FACES system, not Clearview, and the ACLU's April 2026 update identifying a fourteenth publicly known US wrongful arrest does not attribute a vendor in most cases. Legal position: the Seventh Circuit overturned the class settlement in the Illinois BIPA consumer privacy litigation on 13 July 2026. Federal contracting continues. ICE Homeland Security Investigations signed a $9.2m contract in September 2025; US Customs and Border Protection signed a one-year $225,000 contract on 11 February 2026 for tactical targeting operations; and the US Army's 1st Special Forces Command has a contract with a base period running March 2026 to March 2027, with options to 2030 and a total potential value of $339,415.

Source: Company statements; ACLU documentation of wrongful arrest cases; contract reporting · last checked 2026-08-25

Legal basis

Not disclosed in the source; no dedicated statute specific to AI-based analytics identified

Oversight

US Government Accountability Office, which conducted the review this entry is sourced from

A GAO review found DHS law enforcement components using more than 20 categories of detection, observation and monitoring technology, including facial recognition and automatic licence plate readers alongside analytic software, some of which uses AI, plus access to third-party ALPR networks. GAO's own remit is bias-risk oversight, meaning this entry originates from an accountability review rather than a DHS disclosure. The review examined fiscal year 2023 usage; the report was published December 2024.

Source: US Government Accountability Office · last checked 2026-08-25

Legal basis

No dedicated statute. Officers are required to review and take responsibility for the final report

Oversight

Internal department review of each generated report before submission; no external oversight body identified

US Department of Justice research documents both products as operational: body-worn camera audio is transcribed and a generative AI model produces a first draft report, which the officer must review, amend as needed and sign. Both systems work from audio only, not the camera's visual footage, according to the DOJ's own description. This differs in kind from a general-purpose assistant like Microsoft Copilot, which the site already tracks separately: these are purpose-built for report drafting from a fixed audio input, not a general chat tool applied to policing tasks.

Source: US Department of Justice research documentation · last checked 2026-08-25

Legal basis

No dedicated statute in most jurisdictions; typically procured under standard municipal contracting

Oversight

Varies by city; several cities have run independent audits of accuracy and cost-effectiveness

Acoustic sensors distributed across a coverage area analyse sound signatures to detect and locate probable gunfire in real time, alerting police. One of the most widely deployed AI-enabled policing technologies in the US, though this entry was missing from the tracker despite the site's existing gunshot-detection topic tag anticipating it. Accuracy and cost-effectiveness have both been contested in independent city audits in several jurisdictions.

Source: SoundThinking company disclosures and multiple city audits · last checked 2026-08-18

Legal basis

Departmental policy; no statutory basis or external regulator identified

Oversight

Internal supervisory review; use-of-force and problematic language flags require sergeant review

Implemented on 1 February 2024 for half the patrol force as a randomised controlled trial, which is why this deployment rests on peer-reviewed evidence rather than vendor material. Natural language processing runs over the audio of every body-worn camera activation once cameras are docked, producing transcripts and automatic flags for professionalism, de-escalation, use of force and language. Apache Junction Police Department has also been studied. Transcription and annotation accuracy is reported to exceed 90 percent, a vendor-derived figure cited within the peer-reviewed study.

Source: Policing: A Journal of Policy and Practice (Oxford University Press), peer-reviewed · last checked 2026-08-20

Legal basis

Municipal contracting; no dedicated statute

Oversight

LAPD Office of the Inspector General reviewed the deployment

The LAPD was the earliest adopter and co-developer, dropping the system in 2020 citing budget, after an inspector general report found it unclear whether the software affected crime trends. Geolitica has since ceased operations, with SoundThinking acquiring specific intellectual property and hiring the engineering team. A Gizmodo and Markup analysis of millions of predictions found on an unsecured server reported the software disproportionately and often relentlessly targeted low-income communities of colour.

Source: WIRED; Gizmodo and The Markup analysis · last checked 2026-08-20

Legal basis

State v Loomis (Wisconsin Supreme Court) held that a sentencing court may use a COMPAS risk assessment as a relevant factor but that it cannot be determinative, and required cautionary warnings about the instrument's limitations

Oversight

State court rules and case law; no dedicated regulator

A fourth-generation risk and needs instrument combining static criminal history with dynamic criminogenic factors such as social environment and employment, producing general recidivism, violent recidivism and pretrial misconduct risk scores. Designed for correctional case management and rehabilitation planning, its use expanded into judicial sentencing. Broward County, Florida is a widely cited example. The detailed input factor list circulating in secondary sources could not be confirmed against a primary vendor document, and is therefore not stated here as definitive. The legal position, by contrast, rests on the Loomis judgment and is well established.

Source: Harvard Law Review analysis of State v Loomis; California Courts summary · last checked 2026-08-25

Legal basis

Arnold Ventures terms of use plus local court rules and statutes

Oversight

No formal regulator; ongoing independent validation including MDRC evaluation

An actuarial assessment using nine factors drawn from age and criminal history, explicitly excluding any interview, community ties, neighbourhood or marital status. It produces three estimates: failure to appear, new criminal arrest, and new violent criminal arrest while on pretrial release. Judges consider the scores alongside other information when setting release conditions, and scores are provided to the accused, the judicial officer, defence and prosecution. Piloted with selected jurisdictions from 2013 and released publicly in 2018 after five years of testing, built on around 750,000 cases from roughly 300 jurisdictions and validated on more than 500,000. Its deliberate exclusion of socio-economic and neighbourhood factors is a direct design response to bias criticism of earlier instruments.

Source: Advancing Pretrial Policy and Research; MDRC evaluation · last checked 2026-08-20

Legal basis

Not stated

Oversight

Not stated

Used for periods when multiple callers report the same incident, such as a road collision. The AI assesses incoming calls and identifies those relating to an incident already known to the emergency communications centre, allowing human call-takers to prioritise calls requiring intervention. Human staff remain responsible for calls needing emergency intervention. Carbyne and the district describe the technology as having processed more than 3,500 events during a 90-day period and reduced redundant calls by more than 30 percent among the calls it triaged. These are vendor and customer reported figures rather than independently validated efficacy measures. Axon announced a definitive agreement to acquire Carbyne on 4 November 2025 for $625 million, closing in Q1 2026 per Axon's FY2025 shareholder letter. Axon separately acquired Prepared (see the AI Emergency and Non-Emergency Call Triage entry) in late 2025; the two now anchor a combined product line branded Axon 911. This concentrates a striking share of the US emergency-response AI market in one vendor -- Axon now owns Carbyne, Prepared, Fusus, Draft One and Axon body cameras. A scope note: this is an emergency communications district handling police, fire and medical calls, rather than a police force. It is included because control rooms and emergency response fall within this site's subject scope.

Source: Axon press release; Axon investor relations; GovTech · last checked 2026-08-25

Legal basis

Not established; reporting has questioned whether the deployment was reviewed under Seattle's surveillance technology framework

Oversight

Seattle surveillance technology framework, with reporting questioning whether it was applied

Analyses emergency medical 911 calls in real time, providing prompts to dispatchers and assisting decisions about whether particular callers should be routed towards a nurse-staffed service rather than receive an immediate ambulance response. In use since December 2023. This is a more consequential application than transcription, and the distinction matters: the AI output influences emergency response prioritisation rather than merely recording what was said. Documented governance concerns centre on the absence of public disclosure and review under the city's surveillance technology framework, though these derive from investigative reporting rather than a regulator's finding. As with the New Orleans entry, this is a fire department rather than a police force, included because emergency response and control rooms fall within this site's scope. On the department's own effectiveness claim, the two figures it has given do not agree. A 2024 Corti press release quoted SFD Medical Director Michael Sayre crediting the technology with a 50% increase in 911 calls routed to the nurse line; a Fire Department spokesperson later corrected that to 32%. Neither figure has been independently verified. The department also confirmed the system had run without disclosure to callers and without public review, and self-determined that it did not qualify for review under Seattle's 2017 surveillance ordinance.

Source: Seattle Times investigation, via GovTech · last checked 2026-08-24

Legal basis

Federal Aviation Administration waiver permitting flight beyond visual line of sight across much of New York City with no visual observer, up to 200 feet above ground level or within 50 feet of structures

Oversight

Federal Aviation Administration waiver conditions; no dedicated civil liberties oversight identified

A drone is launched to a call for service ahead of officers, with a remote pilot flying the whole mission and no personnel on scene. Live video streams to responding officers and evidence is captured from overhead. Onboard autonomy avoids obstacles, and traffic alerting warns the operator of low-flying crewed aircraft. The FAA waiver is the enabling factor rather than the technology itself, which is why the regulatory position is recorded here in as much detail as the capability. Chula Vista Police Department in California has received authorisation for low-altitude close-proximity emergency operations, and Redmond Police Department in Washington has launched a programme. In March 2026 Skydio announced that 12 public safety agencies had received FAA approval for one pilot to fly up to four drones beyond visual line of sight.

Source: Skydio; Federal Aviation Administration approvals · last checked 2026-08-20

Legal basis

None stated

Oversight

None identified; the ACLU of Massachusetts previously demanded disclosure of the force's Spot testing

The Bomb Squad operates two Spot platforms alongside its Special Tactical Operations Team, sending them to barricaded-suspect and potential hostage incidents, to investigate suspicious packages and to help secure crowded events, feeding situational awareness back to troopers and the incident commander so people do not have to approach on foot. Deployments have included a barricaded-suspect shooting scene in Danvers, Boston's Independence Day celebration and the Boston Marathon. Spot is a teleoperated mobility platform carrying cameras, roughly 75 pounds and costing around 100,000 dollars, not an autonomous decision-making system. The NYPD separately suspended Spot in 2021 after public backlash, then acquired two units in 2023.

Source: Bloomberg; Boston Globe; Police1 · last checked 2026-08-25

Legal basis

Municipal contracting

Oversight

None identified

A self-driving camera-equipped patrol robot deployed in Times Square subway station, accompanied by police officers, as a trial of crime prevention effectiveness. Leased at 9 dollars an hour under a seven-month contract worth 12,250 dollars. The NYPD retired the robot in January after four months deployed and confirmed the pilot was complete. Huntington Park Police Department in California renewed a Knightscope contract for a sixth year in August 2024, though that detail is vendor-sourced and carries lower confidence.

Source: Fox News; Knightscope contract announcement · last checked 2026-08-25

Legal basis

Statutory mandate on the Department of Homeland Security to build an integrated automated entry and exit system matching biographic and biometric records of noncitizens

Oversight

Department of Homeland Security Office of Inspector General

Captures a live photo of the traveller, taken by CBP or an approved airline, airport or cruise partner, and compares it against passport, US visa or other DHS-held photographs. A match verifies identity and the system flags people who may be inadmissible, including visa overstays. On non-match the traveller goes through traditional inspection, and US citizens and eligible foreign nationals may opt out and request a manual document check. CBP told Congress it had processed over 193 million travellers with facial biometrics, confirmed more than 163,000 visa overstays and prevented over 1,500 imposters from entering. The effectiveness evidence is mixed and both sides are recorded here. A DHS Office of Inspector General review found that between May 2019 and September 2021, CBP officers referred around 23,000 travellers aged 14 and over to secondary inspection when photos could not be matched. Separate reporting found CBP scanned over 23 million faces at more than 30 entry points during 2020 without identifying a single imposter.

Source: CBP congressional testimony; DHS Office of Inspector General report OIG-22-48 · last checked 2026-08-20

Legal basis

New Orleans city ordinance restricting police facial recognition; the ACLU alleges the department's use breached it

Oversight

New Orleans City Council; ACLU of Louisiana investigation

Project NOLA is a private non-profit founded in 2009 by a former police officer, based on the University of New Orleans campus, acting as a hub for live video from thousands of cameras on private volunteer properties whose owners pay annual connectivity fees. Around 200 of the most sophisticated cameras run live facial recognition, continuously matching passers-by against hot lists of roughly 250 individuals. An automated voice alerts staff on a match; staff review and relay information to police once an evidentiary threshold is judged met, including unsolicited tips. From early 2023 until April 2025 the police department received real-time facial recognition alerts from a network of over 5,000 cameras, more than 200 facial-recognition equipped, despite city restrictions. The department suspended the real-time alert partnership pending a legal compliance review; the camera network was later reopened to police while facial recognition alerts stayed suspended. The City Council has moved towards an ordinance that would make New Orleans the first US city to formally authorise this form of facial recognition policing. Recorded as paused to reflect that the camera access continues while the facial recognition alerting does not.

Source: NPR; ACLU of Louisiana · last checked 2026-08-20

Legal basis

City anti-violence initiative; commuters and civil rights groups argued mass subway scanning is impractical and unconstitutional

Oversight

Evolv has faced federal investigations into its marketing by the Federal Trade Commission and the Securities and Exchange Commission

A month-long pilot across 20 subway stations, beginning at Fulton Street on 26 July, reporting 2,749 scans. It found zero firearms, produced false positives and detected some knives. The department did not disclose how many passengers opted out or how many officers were needed to run the devices, and no agreement with Evolv had been finalised while the pilot outcomes were being assessed. Recorded as discontinued on the basis that the pilot completed with no continuing deployment established. LA Metro has run a comparable pilot at Union Station.

Source: Associated Press · last checked 2026-08-25

Legal basis

No dedicated statute; deployed without a completed Privacy Impact Assessment

Oversight

DHS OIG audit opened Jan 2026 after Senators Warner and Kaine raised concerns; no completed PIA

Smartphone app letting ICE/CBP officers photograph a person during a street encounter and match against CBP's Traveler Verification Service (same system used for airport processing), with contactless fingerprint capture against IDENT as a fallback if no facial match returns. A 'Super Query' feature reportedly searches databases covering individuals, vehicles, aircraft, vessels and addresses simultaneously. Used over 100,000 times since launch per a January 2026 Illinois/Chicago lawsuit. A Privacy Threshold Analysis states ICE does not offer the option to decline biometric collection, retains photos and fingerprints for 15 years including non-matches, and may collect regardless of citizenship or immigration status. May be extended to 1,000+ state and local 287(g) agencies. Differs from UK operator-initiated facial recognition in three ways: it queries border/travel biometric infrastructure rather than a policing watchlist (so includes people never arrested), adds fingerprint capture as a fallback, and operates federally -- bypassing the 20+ US city and state facial recognition bans. On 5 Feb 2026 the 'ICE Out of My Face Act' was introduced to ban ICE/CBP biometric use; a coalition including EFF and EPIC has demanded DHS shut the app down.

Source: ACLU; Senator Markey follow-up letter to ICE; Immigration Policy Tracking Project · last checked 2026-08-24

Legal basis

No dedicated statute; US courts have given wide scope to surveil prisoner calls for safety/crime prevention; all parties are notified calls are monitored

Oversight

None identified externally

Transcribes prisoner telephone calls at scale using Amazon Transcribe plus keyword and semantic analysis, flagging calls for investigators. Now extended (Aug 2026) into continuous voice-identity tracking, which alerts when a caller's voice does not match the person assigned to the PIN used, and into AI video analysis for facility monitoring. Florida DOC paid $2.5m from Aug 2023 for up to 50 million minutes/year; Georgia DOC monitored 7.5m+ calls by April 2021, flagging 1,612 as showing evidence of criminal activity; company reports 65m+ calls/year across seven states. A Thomson Reuters Foundation investigation found the system used beyond its stated gang/violence rationale, including flagging a call reporting a COVID outbreak and one about unemployment benefits, and reportedly configured to flag Spanish words for 'lawyer' -- raising a legal-privilege concern. Differs from the existing seized-device analysis entry in this category by monitoring live telephone infrastructure continuously and at population scale, including family members who have not been convicted of anything.

Source: Thomson Reuters Foundation investigation; Biometric Update · last checked 2026-08-25

Legal basis

Ordinary municipal procurement; no dedicated statute

Oversight

Governed by individual memoranda of understanding and local retention laws per force

Cloud platform integrating existing camera, ALPR, drone and dispatch feeds into one live operating picture for a police real-time crime centre. Its distinguishing feature, the fususCORE device, lets private camera owners (businesses, schools, households) voluntarily connect cameras to the police platform on customisable terms; some jurisdictions run donor programmes providing free units. Axon states there is no facial recognition in the platform -- a specific, checkable claim. Documented contracts: Providence RI ($750,000, signed June 2024, to at least 2028); Chula Vista CA (Jan 2026 council report, replacing Motorola Solutions -- a second entry for a force already on this tracker for Drone as First Responder); Knoxville TN ('Connect Knoxville', residents/businesses buy fususCORE devices); Lawrence KS. Related: Axon Lightpost, a streetlight-powered ALPR camera that pairs with Fusus, making Axon a direct ALPR competitor to Flock Safety (Axon was previously a Flock investor before the two split).

Source: Chula Vista City Council staff report; Stanford Law CodeX presentation transcript · last checked 2026-08-25

Legal basis

Sole-source federal contracts; no dedicated statute

Oversight

No dedicated regulator; Congressional oversight letter sent April 2026 by Rep. Dan Goldman and colleagues

Three related Palantir systems for ICE: (1) ICM, case management, $139.3m contract from 2022 renewed sole-source, combined deal reportedly exceeding $145m; (2) ImmigrationOS, a ~$30m no-bid 2025 contract modification providing 'near real-time visibility into instances of self-deportation' and managing the 'end to end immigration lifecycle'; (3) ELITE, an app generating deportation-target dossiers with a confidence score (out of 100) on a target's address. In M-J-M-A v Wamsley (D. Oregon), ICE officers confirmed under oath on 2 Dec 2025 that ELITE was used to identify arrest targets during Oregon raids driven by daily arrest quotas -- sworn testimony, stronger evidence than surrounding reporting. 404 Media reported ELITE drawing on HHS/Medicaid data; Palantir disputes this characterisation as misleading. A CMS-ICE data-sharing agreement exists; a preliminary injunction (Aug 2025) briefly blocked sharing in 20 states before a Dec 2025 ruling allowed some sharing to resume pending the case. Same vendor as this site's existing Palantir Gotham entry for German state police, where a Federal Constitutional Court judgment constrains automated analysis -- the US deployment has essentially the inverse legal position: no court constraint, sole-source contracts, and an operational target list rather than an investigative link chart.

Source: ACLU; American Immigration Council; court filings in M-J-M-A v Wamsley · last checked 2026-08-25

Legal basis

Ordinary municipal procurement; no dedicated statute

Oversight

None identified externally; Antioch's staff report frames dispatchers as retaining prioritisation and dispatch decisions

Assistive AI layer for 911 call centres: transcribes calls live, translates in real time via synthetic voice, and extracts addresses/incident details automatically. Reports 1,000+ agency partnerships covering ~100 million people; combined with Carbyne, Axon states the pair process 80m+ 911 calls/year. Trade publication Officer.com reports (Aug 2026, citing Axon's 911-division CEO) that in Fort Worth TX the platform handles non-emergency calls fully autonomously, thousands per month with no human call-taker -- a categorical step beyond every other control-room entry on this tracker, which all describe AI assisting a human who retains the decision. This specific claim is vendor-sourced (a company CEO in trade press, not a Fort Worth city source) and is recorded with that caveat; if independently confirmed it would be the first fully automated call-handling deployment found in this research. Contrasts directly with this site's existing UK 101-triage entry, which records the UK government's position that it 'does not support police use of AI to answer 999 calls'. Confirmed municipal procurement: Antioch CA (March 2026 sole-source staff report) and Alameda County Sheriff's Office.

Source: Officer.com; Antioch City Council staff report via Contra Costa News · last checked 2026-08-25