Last updated:
📊 What 600 Million ALPR Audit-Log Rows Show
Every released licence-plate-reader audit says the same things: the cameras read hundreds of millions of plates per agency, the searches are mostly by someone else, the reason field is mostly noise, immigration searches are wherever anyone has looked, hits are a fraction of a percent, and misuse appears by the dozen once anyone audits. The state audits that say 'no exceptions' check settings, not searches. 799 findings, with their sources, are in the companion dataset.
Quick answer: The audit logs say the same things in every jurisdiction that has released them. The cameras read plates at a scale nobody budgeted for: 830 million a year in Maryland, 639 million in Oakland alone. Most searches of a town’s cameras are not by the town. The reason field is mostly empty of meaning. Immigration searches are in the logs wherever anyone has looked. Hits are a fraction of a percent of reads, and arrests a fraction of that. Misuse, when someone finally audits, turns up by the dozen. The compliance audits states require say “no exceptions” because they check settings, not searches. Every number below is in our dataset with its source and the sentence it came from.
What an audit log is, and why there are now 799 findings and 600 million rows
Every Flock search writes a row: who searched, from which agency, when, with what reason text, which case number if any, and how many networks and cameras the search reached. Agencies can export two versions. The organisation audit lists their own users’ searches. The network audit lists every search that touched their cameras, including searches by other agencies through state or national sharing. Both are public records in most states, and since 2025 people have been requesting them in volume.
Have I Been Flocked, a volunteer project, now catalogues 3,231 released files from 151 agencies in 29 states, about 610 million raw rows that de-duplicate to 291 million unique searches Have I Been Flocked (read 2 Nov 2026). EFF’s own sample is 12 million searches by 3,900 agencies from December 2024 to October 2025 EFF 2025 in review (Dec 2025). USA Today analysed 75 million searches from 120-plus agencies in 24 states USA Today, Frequent Flockers (4 Sep 2026). The FlockWatch corpus, released 31 October 2026, de-duplicates 887,928,694 released rows to 153,953,051 unique searches by 6,696 agencies FlockWatch corpus 2026-10-31. On top of the logs sit the official records. Statutory reports in Maryland, Vermont, Illinois, Nebraska and North Carolina. Inspector-general and auditor reports in Los Angeles, Seattle, Austin and San Mateo County. Privacy-commission reports in Oakland, Berkeley and San Diego. Court filings in Norfolk, San Jose and Westchester County. Our dataset holds 799 findings from 475 sources, each read at the source. This article is what they add up to.
Finding 1: the scale is in the hundreds of millions of reads per agency
Maryland’s fusion centre retained 830,097,779 plate reads in 2023 from 698 reader units, up from 450 million in 2018. It fielded 5,310 requests for that data, 684 of them federal Maryland MCAC statutory LPR report for 2023 (28 Feb 2024). By 2025 local reporters counted at least 4.8 billion scans across 26 Maryland agencies Capital News Service Maryland (27 Oct 2026). Oakland’s 290 Flock cameras read 638,747,333 plates in 2025, more than three times the 189 million of July to December 2024 OPD 2025 ALPR annual report, Privacy Advisory Commission packet (4 Jun 2026). LAPD’s cameras read 210,568,103 plates in August and September 2025 alone LAPD Office of the Inspector General, BPC 26-184 (10 Jul 2026). Westchester County, New York, logged 264,303,687 reads in 2024 and holds about 1.6 billion, kept for at least two years Policing Project, Umemoto v. Westchester County (9 Jun 2026). Norfolk’s 175 cameras logged 41,148,215 captures in a 29-day window IJ opening brief, Schmidt v. Norfolk (Apr 2026). North Carolina’s state pilot captured about 153 million plates from 140 camera locations in eight months; New Hanover County’s 19 cameras alone took 35.5 million NC SBI pilot report to the legislature (7 Apr 2026).
Smaller places scale the same way. Everett, Washington, read 8.5 million plates in April 2026 Everett PD Flock report, April 2026. Marion County, Oregon, read 3.07 million in August 2026 from 11 cameras Marion County SO ALPR monthly report, August 2026. East Palo Alto’s 25 cameras read between 3.2 and 5.3 million a month. Iowa’s portals averaged 10,177 detections per camera per month ACLU of Iowa / University of Iowa Technology Law Clinic (10 Dec 2025). A plate reader is not a camera that watches a corner. It is a sensor that records everyone, and the records are the product.
Finding 2: the cameras are mostly searched by someone else
This is the finding every log repeats. Dunwoody, Georgia’s 2025 network audit holds 23,426,238 rows from about 3,500 organisations; its own officers account for 54,148 Have I Been Flocked, Dunwoody GA data update (23 Mar 2026). Five of the twelve monthly files were truncated at the export limit, so the real gap is wider. Coconino County, Arizona, found its cameras searched more than three million times by outside agencies from January to July 2026, against 3,501 of its own Williams-Grand Canyon News (Aug 2026). That is about 17 a day inside and more than 16,000 a day outside. Syracuse’s data was searched nearly 4.4 million times in a year by police across the country; fewer than 500 of those searches were about Syracuse alone Central Current (4 Jul 2025). Thornton, Colorado’s December 2025 network audit stopped at 1,048,576 rows, which is exactly Excel’s limit Thornton PD ALPR memo (1 Apr 2026). The department ran 46,230 searches in the whole of 2025 and has since revoked more than 2,100 agencies’ access. Woodburn, Oregon’s 24 cameras were searched 3.3 million times by 4,734 agencies in eleven months; 306 searches targeted Woodburn specifically Woodburn Independent (26 Feb 2026). Lexington, Kentucky’s 125 cameras took more than 7 million searches in the first seven months of 2026; 96% came from out-of-state or federal agencies, and Lexington’s own police ran 63,610 WKYT Investigates (15 Sep 2026). Weber County, Utah’s ten cameras were searched 5.1 million times by 3,343 agencies, almost 97% out of state, Houston alone more than a third Utah News Dispatch (14 Aug 2026). Monongalia County, West Virginia’s cameras were searched 3,049,590 times by 3,918 organisations in nine months while the county’s portal advertised 85 searches in 30 days DeFlock Monongalia (8 Oct 2026). McDonough County, Illinois’ sheriff ran 575 searches in 13 months; other agencies ran 6,158,356 on his cameras, 10,710 for each of his Have I Been Flocked, McDonough County IL (19 Jun 2026). Adams County, Washington, population 19,000: the sheriff’s own searches were 0.08% of the 2.1 million on his network, and none of the top fifteen searchers was in Washington Have I Been Flocked, Adams County WA (22 Jun 2026). In Erie, Colorado, Houston PD alone was 58.5% of the month’s searches Have I Been Flocked, Erie CO (7 Aug 2026). Routt County, Colorado’s 13 cameras were included in more than 8 million searches in twenty months, about 396,000 a month in 2026 Yampa Valley Bugle (1 Sep 2026). North Carolina’s campus networks are the extreme case: UNC Charlotte’s cameras took 3,775,134 searches in three months from 3,806 agencies, 94% outside the state, and the campus police ran about 228 DeFlock ILM (17 Oct 2026).
The state networks show the same shape. Texas DPS’s network took roughly 615,000 searches in June 2026, one every four seconds, from 3,075 agencies Texas Tribune (4 Sep 2026). 2,679 of those agencies were outside Texas, and 48% of queries came from out of state. One school district’s eight cameras in Alvin, Texas, were searched more than 733,000 times in a month by more than 3,100 agencies The 74 / The Guardian (10 Feb 2026). In North Carolina, about 90% of 3.9 million searches touching five small networks came from out of state DeFlock ILM methodology (2026). In Kyle, Texas, outside agencies ran 1.5 million searches against 2,888 by Kyle’s own police FOX 7 Austin (13 Sep 2026). Colleyville’s transparency page said 44 searches in 30 days; its released logs averaged more than 700,000 a month Current Revolt (22 Oct 2026).
California shows what happens when the sharing is illegal rather than merely invisible. SFPD’s data was searched more than 1.6 million times by out-of-state agencies in seven months SF Standard (8 Sep 2025). Ventura County’s was searched 364,000 times in a single month after a National Lookup setting was enabled without the agency’s knowledge Ventura County Sheriff's Office statement (27 Feb 2026). Marin County’s sheriff’s data was searched 254,131 times in November 2024 while shared with out-of-state and federal agencies EFF on Marin County (1 Oct 2026). Mountain View’s was searched about 600,000 times by more than 250 unauthorised agencies in a year Mountain View Voice (30 Jan 2026). Lynnwood, Washington, turned nationwide lookup on for eleven days and logged more than 100,000 out-of-state searches Everett Herald (1 Nov 2025).
Across the 27 agencies in our index that released both file types, the median ratio of network rows to own-search rows is 556 to 1. Riverside County is at 9 to 1; Barrington, Illinois, at 8,378 to 1. The searching side is concentrated. Of 309 million searches attributed to a named agency, the top ten account for 17%. The leaders are Houston TX PD (14.0M), Riverside County CA SO (10.3M), Dallas TX PD (6.4M), California Highway Patrol (5.7M), San Francisco CA PD (4.7M). Those agencies run nationwide searches, so each query lands in hundreds of other logs; the number measures reach as much as volume.
Finding 3: the reason field is mostly noise
Flock requires a reason for every search and, since August 2026, is moving to require a case number. The logs show what a free-text field does. In Dunwoody, 1.3% of the department’s own January 2025 searches carried a case number; by November, after scrutiny, 26.9% did Have I Been Flocked, Dunwoody GA data update (23 Mar 2026). In Thornton, about 8% of 2025 searches included a case or call number Thornton PD ALPR memo (1 Apr 2026). In Joplin, Missouri, an audit found 85% of searches had one-word justifications and under 3% linked to a criminal case Washington Post via ADN (2 Aug 2026). Boston’s five-month pilot logged 11,004 searches, about half with a case number WBUR (14 Sep 2026). In Norfolk, the three most common reasons officers typed were “cj”, “stolen” and “inv”, nearly 75,000 of 230,000 searches; others read “protestor identification” and “la la la la” IJ opening brief, Schmidt v. Norfolk (Apr 2026). In Fall River, Massachusetts, under 3% of searches from 2021 to 2025 carried a case number; one officer typed “99” as his reason more than a thousand times The New Bedford Light (6 Oct 2026). Brunswick County, North Carolina’s sheriff ran 83,232 searches, 90.1% with no case number DeFlock ILM (7 Sep 2026). Across the whole FlockWatch corpus of 154 million de-duplicated searches, 84.9% have no case number and the reason field holds 2,164,768 distinct strings FlockWatch corpus 2026-10-31. In Monongalia County, 81.38% of 3 million searches carried no case-specific content and 362,452 were labelled “Traffic Infraction” DeFlock Monongalia (8 Oct 2026). Santa Clara County’s sheriff released 23,860 searches with no reason and no case number on any of them Have I Been Flocked, Santa Clara County SO (18 Jun 2026). Napa County’s most common reason was “work”, 25,022 times Have I Been Flocked, Napa County SO (3 Oct 2026). In Jefferson County, Missouri, one operator searched one plate 239 times in three hours under “Traffic Infraction” Have I Been Flocked, Jefferson County MO SO (3 Oct 2026). The exceptions prove the point: Springfield, Massachusetts, logged a case number on 100% of its own searches, and Prescott, Arizona, went from 15.8% to 100% the day it made the field mandatory Have I Been Flocked, Prescott AZ (3 Oct 2026). Pryor Creek, Oklahoma: 83% of 2,283 searches with no case number The Sooner Sentinel (30 Oct 2026). Bexar County: placeholder case numbers such as “12345” News 4 San Antonio (12 Oct 2026). Across about 200,000 Virginia searches in the first four months of the state’s new law, more than 1,800 used “suspicious” or “suspect” and nothing else VCIJ at WHRO (2 Apr 2026).
Nationally, EFF found more than 6,300 searches by more than 30 agencies with the reason “TBD”, 1,954 from one Alabama department, alongside “LOL”, “LMAO” and keyboard mashing EFF (Sep 2026). In Wisconsin, “investigation” alone was the most common reason statewide Wisconsin Examiner (6 Aug 2025). Flock’s FreeForm feature lets an officer search by description instead of plate. Of 3,217 such searches across 124 agencies, 3,184 were allowed, 19 blocked and 14 warned Footnote4a (9 Mar 2026). One Houston query for a white car reached 53,017 devices. In New Castle, Pennsylvania, seven officers kept private hot lists of 99 plates, 95 without a case number and 28 set never to expire Footnote4a (18 May 2026).
The agencies that score well are the ones that audit. Berkeley reviewed every access record for the first half of 2025 and found 99.81% of external searches compliant Berkeley PD 2025 surveillance technology report (5 Nov 2025). Oakland sampled 500 records and found 98% carried a report number OPD 2025 ALPR annual report, Privacy Advisory Commission packet (4 Jun 2026). Alexandria flagged 56 searches in three years of 30-day audits and found no misuse City of Alexandria LPR dashboard (30 Oct 2026). Those are the exceptions. San Mateo PD had never conducted a search-activity audit in eight years of ALPR use sm-alpr findings (Jul 2026). Norfolk first downloaded its audit files in February 2025 and first reviewed them in May, after being sued IJ opening brief, Schmidt v. Norfolk (Apr 2026). Prairie Grove, Illinois, logged 2,993 searches by two people, seven in eight labelled “training” or “test”, until one of them was arrested for stalking his ex Footnote4a (19 Jun 2026).
The reason field matters because it is the only thing an auditor can read. A search logged as “99” is unreviewable, and an agency that accepts it is not auditing.
Finding 4: immigration searches are wherever anyone has looked
Dayton’s released logs show 7,150 outside accesses for immigration-related reasons since July 2023: 22 in 2023, 1,024 in 2024, 6,104 in 2025. The most common reason was simply “ICE”, 1,776 times; U.S. Border Patrol accounted for 3,703 Dayton Daily News (26 Jun 2026). Columbus audited 19,664,308 searches of its network over two and a half years: ICE itself ran zero, Border Patrol and HSI ran 3,962, other federal agencies 79,236, and 15,577 searches by any agency carried a possibly immigration-related reason WSYX on the Columbus audit (14 Jul 2026). Syracuse: 2,097 of 4.4 million Central Current (4 Jul 2025). Denver: more than 1,400 immigration-related national searches that included its cameras, about 690 of them after January 2025 Colorado Newsline via KUNC (10 Aug 2025). Santa Cruz: nearly 4,000 searches by 32 California agencies with immigration-related terms, sharing that state law bars Santa Cruz Local (9 Dec 2025). Cleveland Heights: 1,783 across 20 cameras in a year cleveland.com (Jul 2026). Shaker Heights: 273 of 694,589 searches reviewed, every one by an outside jurisdiction City of Shaker Heights (8 May 2026). Shelton, Washington: 943 outside searches with a strong immigration connection in 2.67 million entries KIRO 7 (11 Oct 2026). Woodburn: 384 direct searches by DHS and Border Patrol in fifteen weeks of a pilot, plus 261 immigration-related searches by out-of-state police Woodburn Independent (26 Feb 2026). Bend: 279 federal or ICE-related queries in the first three weeks its cameras were live The Source Weekly (6 May 2026). Alvin ISD: 620 immigration-related searches of school cameras by 30 agencies in one month The 74 / The Guardian (10 Feb 2026). Routt County, Colorado: nearly 1,300 by more than 50 agencies Yampa Valley Bugle (1 Sep 2026). Twin Falls, Idaho: at least 631 in three months Idaho Capital Sun via Route Fifty (11 Aug 2026). Florida’s Treasure Coast: 216 in the first quarter of 2026 against zero in the same quarter of 2025 TCPalm (1 Jun 2026). Danville, Illinois: more than 4,000 nation- and statewide lookups tied to federal requests 404 Media (27 May 2025).
The searching agencies are not only federal. Texas DPS’s June 2026 log had 111 searches citing HSI, CBP or immigration, including “ICE deportation warrant” Texas Tribune (4 Sep 2026). Lexington, Kentucky, ran 83 searches at the request of ICE and HSI investigators in six months Spectrum News 1 Kentucky (21 Sep 2026). The Florida Highway Patrol ran more than 12,000 nationwide searches in seven months of 2025, with explicitly immigration-related reasons among them ACLU of Massachusetts letter (Jan 2026); Florida’s wildlife police ran 38 searches for ICE in January 2026, each reaching more than 5,000 networks 404 Media (6 Apr 2026). In the released logs, the most recent 10,000 searches whose reason names immigration cover only 25 February to 6 October 2026, led by Jefferson Parish, Louisiana (2,021), the Florida Highway Patrol (1,643) and Florida Fish and Wildlife (588) Have I Been Flocked (read 2 Nov 2026). Illinois’ Secretary of State found 46 out-of-state agencies searching Illinois data illegally and 262 immigration-related searches of one suburb’s system Capitol News Illinois (12 Jun 2025). Washington’s review of 31 agencies found eight sharing directly with Border Patrol UW Center for Human Rights (21 Oct 2025).
Where the answer was “few”, it still was not zero. Ventura County: 299 immigration-citing justifications, 0.08% of searches Ventura County Sheriff's Office statement (27 Feb 2026). Los Altos: 148 out-of-state searches citing immigration or ICE in two years Los Altos PD community message (9 Mar 2026). Berkeley: one “ICE” and two “CBP” in six months Berkeley PD 2025 surveillance technology report (5 Nov 2025). Palo Alto reported none Palo Alto PD chief's statement (21 Apr 2026); LAPD’s July 2025 term inspection found none LAPD Office of the Inspector General, BPC 26-184 (10 Jul 2026). Maryland’s fusion centre, which does not use Flock, released data to federal agencies 591 times in 2023 on 684 requests Maryland MCAC statutory LPR report for 2023 (28 Feb 2024). The dataset’s searcher table lists federal agencies that appear by name; searches run through a local login do not, so that list is a floor.
Finding 5: hits are a fraction of a percent, arrests a fraction of that
The agency reports now give denominators. Oakland: 638.7 million reads, 1,099,837 hot-list alerts (0.17%), 162 arrests; the stolen-vehicle and stolen-plate alerts, 653,566 of the total, were switched off because of the volume OPD 2025 ALPR annual report, Privacy Advisory Commission packet (4 Jun 2026). LAPD: 210.6 million reads, 50,180 hits on 5,911 unique plates, 68 stops, 74 arrests, 337 vehicles recovered; officers confirmed 161 alerts as accurate matches on vehicles that turned out not to be stolen LAPD Office of the Inspector General, BPC 26-184 (10 Jul 2026). Norfolk: fewer than 2% of detections trigger any alert IJ testimony to the Senate Judiciary Committee (23 Sep 2026). San Jose: roughly a quarter of one percent of 360 million photographs triggered a hot-list alert Tan v. City of San Jose complaint (15 Apr 2026). Nashville’s 2023 pilot: 71 million plates scanned, 1,316 verified hits, 63 arrests Nashville Scene (Aug 2023). New Hanover County, North Carolina: 2,980,082 searches of its network against 43 logged arrests, about one per 69,000 searches DeFlock ILM (11 Sep 2026). San Diego: more than 244,000 searches in 2025, 269 arrests SDPD 2025 annual surveillance report (Feb 2026). Iowa: 13 hits per 1,000 detections across the state’s portals ACLU of Iowa / University of Iowa Technology Law Clinic (10 Dec 2025). Austin: 75 million scans, 13,122 hits, 286 stops Austin Office of the City Auditor (May 2025). Seattle’s inspector general: unverified hits about 0.1% of valid reads, about one misread per five hits, and two erroneous high-risk stops in two years Seattle OIG ALPR review (29 Sep 2023). Vermont in 2019: 136,983 reads, 3,196 alerts, 8 enforcement actions Vermont DPS 2019 ALPR report (Jan 2020). Illinois State Police’s expressway cameras report 3.38 billion detections and 94.8 million “criminal database hits”, a 2.8% rate that counts matches, not stops Illinois State Police FY26 Expressway Camera Act report (Jul 2026).
The error side is now documented too. Roseville, California, reviewed 1,427 stolen-vehicle and felony alerts and found misread plates in 1,011, a failure rate above 70% Whitson v. Flock Group complaint (11 Sep 2026). The Institute for Justice counts at least 33 innocent motorists stopped, detained or jailed on ALPR errors since 2018 IJ testimony to the Senate Judiciary Committee (23 Sep 2026). During a Wisconsin Amber Alert, agencies in 23 states ran 3,466 searches for two plates, more than 2,000 of them for the wrong plate, and the suspect was found by a truck-stop tip Footnote4a (28 Dec 2025).
These are not arguments that the cameras never work. They are the denominators that vendor case studies leave out.
Finding 6: when someone finally audits, misuse turns up by the dozen
The Institute for Justice catalogues more than 240 abuse incidents nationwide IJ testimony to the Senate Judiciary Committee (23 Sep 2026). The Washington Post counted at least 50 officers charged or accused, 46 of them using Flock, 26 targeting wives, girlfriends, exes or women the officer wanted to meet Washington Post via ADN (2 Aug 2026). Georgia alone has more than 40 people accused of misusing Flock since November 2025 FOX 5 Atlanta (1 Sep 2026); CNN counts at least 85 cases nationally since 2021 CNN via KQ2 (1 Nov 2026). USA Today’s review of 75 million searches flagged 240 users for persistently searching the same plates and sent the names to 67 agencies, which led to at least 19 employees arrested, fired, put on leave or investigated within two months USA Today (4 Sep 2026). Indianapolis charged five officers in September 2026 after its own audit; one had run about 3,759 searches in ten months WFYI (24 Sep 2026). The individual cases are what an audit log looks like when read. Savannah audited 39,000 searches this year, flagged 127, found 34 unjustified, fired six employees and arrested four WTOC (24 Aug 2026). A part-time dispatcher in Highland Heights, Ohio, ran 20,292 searches in thirteen months, one plate 5,369 times KY3 (21 Aug 2026). Colleton County, South Carolina, fired a second lieutenant for 62 unauthorised searches after the first was found with 2,700 WRDW (31 Aug 2026). A Mauldin, South Carolina, officer searched his ex-partner’s vehicle 166 times in five months FOX Carolina (14 Aug 2026). A former Holiday Hills, Illinois, chief was charged over nearly 250 unauthorised Flock searches and 31 LEADS lookups Chicago Sun-Times (20 Jul 2026). A Milwaukee officer searched the plates of current and former partners 175 times in a year Milwaukee Fire and Police Commission letter (25 Aug 2026). Hall County, Georgia’s audit produced three resignations FOX 5 Atlanta (1 Sep 2026); Milwaukee’s monthly audit flagged 24 outliers in April 2026 FOX6 (28 May 2026); Atlanta found 0.07% of its 2026 searches questionable CNN via KQ2 (1 Nov 2026).
Lexington, Kentucky’s quarterly self-audits are the counter-example: 10,000 to 15,000 searches a month, every one “listed a reason”, no misuse ever reported, an independent sample that shrank from 1% of entries to a flat 60 a quarter, and 185 Audit Assistance alerts in one quarter all cleared as active investigations Lexington PD quarterly Flock audit, Q2 2026. In the same months, outside agencies ran 7 million searches on Lexington’s cameras that the memos never mention. Two things the cases share. Almost every one surfaced only after a records request, a lawsuit or Flock’s own August 2026 audit tool, not from the quarterly sample audits most policies prescribe. And the sample audits missed them: Shively, Kentucky’s ten-searches-per-quarter sample missed more than 2,000 searches by one officer tracking his ex-partner Spectrum News 1 (31 Aug 2026); Norfolk’s oversight had found two instances of misuse in two years IJ opening brief, Schmidt v. Norfolk (Apr 2026). The misuse rate is a function of how hard anyone looks.
Finding 7: the audits states require say “no exceptions”, because they check settings
New Jersey audits every agency every year: 517 certified for 2023, 523 for 2024 and 2025, two significant violations in each of the first two years NJ OAG via WRNJ (28 Jul 2025). Minnesota’s biennial audits test seven controls and report compliance; the better ones now include counts, such as the Paul Bunyan task force’s 1,225,448 reads and 4,941 hits in 30 days Rampart Audit, Paul Bunyan Drug Task Force (28 Nov 2025) and St. Cloud’s 1,053,902 mobile reads Rampart Audit, St. Cloud PD (29 Apr 2026), but none reads the reason field Maple Grove MN biennial ALPR audit (22 May 2026). Maryland’s fusion centre audited 65 of 5,310 requests in 2023 and found all accurate Maryland MCAC statutory LPR report for 2023 (28 Feb 2024). Seattle’s inspector general found all 89 sampled uses conformed to policy Seattle OIG ALPR review (29 Sep 2023). Virginia’s Crime Commission surveyed 159 agencies: 33 retained data beyond the statutory 21 days, 20 gave out-of-state agencies continuous access, nine gave it to federal agencies Virginia State Crime Commission ALPR update (21 Jan 2026). San Mateo County’s grand jury rated 17 agencies’ audits: seven strong, four partial, two weak, four with no audit at all San Mateo County Civil Grand Jury (3 Aug 2026). In Iowa, ten agencies claimed their audit logs were confidential; four produced them ACLU of Iowa / University of Iowa Technology Law Clinic (10 Dec 2025).
Flock itself told the Senate in September 2026 that it keeps no consolidated count of misuse Flock Safety letter to Sen. Hawley (8 Sep 2026). These audits test whether a retention setting is 60 days and whether a sharing policy exists. They do not read the reason field, and a reason-field audit is what finds the dispatcher with 20,292 searches. New Jersey’s 2025 audit found no significant violation in the year a Fall River officer typed “99” a thousand times in the state next door. Both can be true, because they measure different things.
Finding 8: the logs themselves are not stable
Footnote4a, which downloads the same portal logs daily, found that 2.6% to 7.1% of West Des Moines’ records changed between snapshots, and 3.9% to 8.2% of Johnston’s Footnote4a (8 Apr 2026). About 80% of Santa Cruz’s search logs went missing after August 2025, when daily searches fell from 7,000–15,000 to almost nothing Footnote4a (9 Dec 2025). Harris County’s 11.4 million searches resolved to 11,745 unique plates after Flock altered its network logs Have I Been Flocked, Harris County TX (2 Feb 2026). Avondale, Arizona’s network files skip 242 of 823 days Have I Been Flocked, Avondale AZ (3 Oct 2026); Roseville, California, released 208,915 records with every name, plate, reason and case number blank Have I Been Flocked, Roseville CA (3 Oct 2026); Riverside County’s 22 million rows say “Investigation” on 98.8% of them because the export dropped the columns that would say more Have I Been Flocked, Riverside County SO (4 Oct 2026). Network volumes also collapse without explanation: Rogers, Arkansas, fell 85% in a day on 25 March 2026 and Woodbury, Minnesota, lost 90% of its searching agencies in June 2025, consistent with Flock restricting nationwide sharing Have I Been Flocked, Woodbury MN (18 Jun 2026). Thornton’s, Dunwoody’s and North Carolina’s exports were truncated at spreadsheet limits. San Diego’s 2024 report says “more than 140,000” searches in its text and 10,536 in a table SDPD revised 2024 ALPR annual report (23 Oct 2025). Senators Wyden and Krishnamoorthi told the FTC at least 35 Flock customer accounts had stolen passwords Wyden and Krishnamoorthi letter to the FTC (3 Nov 2025). An audit log is evidence only if it is complete and unchanged; several of these were neither.
How to use this
- Request both audits. The organisation audit shows what your officers do; the network audit shows who is using your cameras. Ask for a date range, the data-sharing settings and the search policy together. Templates are in the records-request guide.
- Count five things. Share of searches by outside agencies; share with a case number; reasons that are a single word or no word; any reason containing ICE, CBP, HSI, Border Patrol, immigration, or a protest name; and reads against hits against arrests. Those numbers are the whole argument.
- Compare with the transparency portal. If the portal says 44 and the log says 700,000, that gap is the story, and it is one your council can see without reading a spreadsheet.
- Check for truncation. A file of exactly 1,048,576 or 2,097,151 rows was cut off by the export, not by reality.
- Send the file to Have I Been Flocked. It is the only place the released logs are being joined, and every added file makes the searcher table in our dataset more complete.
Frequently asked questions
What do Flock audit logs show?
The same things everywhere they have been released: plate reads in the hundreds of millions per large agency (Oakland 639 million in 2025, LAPD 211 million in two months); most searches of a local network come from other agencies (Dunwoody GA 23.4 million outside rows against 54,148 of its own); the reason field is mostly meaningless (Dunwoody 1.3% with a case number in January 2025, Joplin 85% one-word reasons); immigration-related searches wherever anyone looked (Dayton 7,150, Columbus 15,577 possibly related of 19.7 million); and hit rates below one percent of reads.
How many Flock searches have been made public?
As of 2 November 2026, Have I Been Flocked catalogues about 3,200 released audit files from about 150 agencies in 29 states, roughly 610 million raw rows that de-duplicate to 291 million unique searches. EFF separately analysed 12 million searches by 3,900 agencies. Our dataset adds findings from hundreds of statutory reports, auditor reports, court filings and press analyses.
Why does a small town's audit log have millions of rows?
Because the network audit records every search that touched the town's cameras, and a nationwide search by any agency touches thousands of networks at once. Coconino County, Arizona, ran 3,501 searches of its own in seven months while outside agencies ran more than three million on its cameras. Woodburn, Oregon's 24 cameras were searched 3.3 million times by 4,734 agencies in eleven months.
What is the hit rate for licence-plate readers?
Below one percent of reads where agencies reported both. Oakland: 1,099,837 alerts on 638.7 million reads in 2025, 0.17%, and 162 arrests. LAPD: 50,180 hits on 210.6 million reads in two months, 74 arrests. Norfolk: fewer than 2% of detections trigger any alert. San Jose: about a quarter of one percent. Austin: 13,122 hits and 286 stops on 75 million scans. Alerts include stolen plates and expired registrations, so stops and arrests are far rarer than alerts.
Do state ALPR audits find misuse?
Rarely, because they test settings rather than searches. New Jersey's annual audit certified 523 agencies for 2025 with no significant violation; Minnesota's biennial audits check seven controls; Maryland's fusion centre audited 65 of 5,310 requests and found all accurate. Misuse has been found by reading the reason field and case numbers in released logs: Savannah flagged 127 of 39,000 searches and arrested four employees; a Highland Heights, Ohio dispatcher ran 20,292 searches before anyone looked.
Were immigration agencies searching local cameras?
Yes, in every state where the logs were read for it. Dayton's logs show 7,150 immigration-related outside accesses since 2023, 3,703 by Border Patrol. Columbus found 3,962 Border Patrol and HSI searches and 15,577 possibly immigration-related searches, with ICE itself at zero. Syracuse 2,097, Denver 1,400-plus, Santa Cruz about 4,000 by California agencies, Shelton WA 943, Bend 279 in three weeks. Searches run through a local login are attributed to the local agency, so named federal entries are a floor.
How many officers have been caught misusing ALPR data?
The Institute for Justice catalogues more than 240 abuse incidents and at least 33 innocent motorists wrongly stopped or jailed. The Washington Post counted at least 50 officers charged or accused by August 2026, 46 using Flock, 26 targeting partners, exes or women they wanted to meet. Georgia alone has more than 40 people accused since November 2025. Most cases surfaced through records requests, lawsuits or Flock's own audit tool rather than routine sample audits.
How do I check whether my plate was searched?
Have I Been Flocked lets you enter a plate and see whether it appears in any released audit log, with the agency and the stated reason. Coverage is only as good as the logs released so far, and can lag by months. To see your own agency's logs, file a records request for its network and organisation audits.