ALPR Threat Brief / Prepared for Kendall County & Boerne, Texas / August 2026
Hypothetical — grounded in documented failures

You can reissue a credit card.
You cannot reissue everywhere you have been.

A national license plate reader network holds something no other breached database holds: the physical movements of ordinary people, timestamped and mapped, going back years. Here is what a bad actor does with that file the day it leaks.

Scan stream · illustrative capturing
Each row is one vehicle, one camera, one moment. Multiply by 20 billion per month.

The size of the file

This is not a niche system. It is one of the largest continuously-updated location datasets on American civilians, and it is assembled by a private vendor on behalf of thousands of separate police departments — each with its own login discipline.

119,000 ALPR cameras mapped by the crowdsourced DeFlock project; ~83% are Flock. Crowdsourcing undercounts — Flock itself claimed 120,000+ in Aug 2026. Fox News, Jul 2026
20B+ Vehicle scans recorded per month across the network. Company figures
6,000+ Communities across 49 states feeding into and searching the same shared pool. Stateline, Aug 2026
7 days New default retention as of Aug 2026, cut from 30. But Flock’s own Evidence Policy still reads “thirty (30) calendar days,” customers may set longer, and “Evidence Mode” preserves searched data past the window. Flock, Aug 2026

What one record actually contains

People picture a photo of a license plate. Below is the record as Flock itself defines it — drawn from the company’s own LPR Policy, its published v3 API schema, and court and public-records disclosures. Each line is tagged with where the claim comes from, because the three tiers do not always agree.

ContractFlock’s LPR Policy — binding language, not marketing APIField names from the published Flock API v3 schema RecordCourt rulings and public-records releases ClaimFlock marketing only — no contract or independent confirmation
One plate read, as the API returns it
id API4171ae2b-843f-405b-922d-b1e89921f24f
capturedAt API2026-08-19T12:11:46Z
plate API7XKR412  confidence 0.94
state APITX  confidence 0.88
device APICAM-4471 · type: camera
device.location API29.7947, -98.7320 · Oak & 12th
network API“Kendall County SO”
link APIshare.flocksafety.com/lookup/…
Held alongside it
Images Recordplate image + vehicle image (LPR Policy). In Washington litigation, Flock cameras were described as capturing 6 to 12 timestamped images of every car that passes.
Vehicle characteristics Contract“vehicle type, license plate information, roof racks, etc.” — the binding text names only color, make, type and roof racks.
Vehicle Signature® Claimbody type, damage, aftermarket wheels, bumper stickers, decals, temporary or missing plate — and matching “even when the vehicle does not have a readable plate.” Flock marketing only.
Query audit log Contractusername, date, time, purpose of query, plate and other search elements. Exports obtained by records request also carry org name, camera count, time frame, filters, search type, text prompt, and moderation result.
Retention ContractThe LPR Policy says 7-day rolling delete by default. Flock’s Evidence Policy, live the same day, still says “thirty (30) calendar days.” Ask which one governs your contract.
1

The location is the camera’s, not the car’s — and that is enough. Coordinates live on the device record, not on each read. A fixed camera’s position is known to four decimal places, so the read still places a specific vehicle at a specific corner at a specific second.

2

Every read carries a network name. That single field is the sharing mechanism. It is how a plate read taken in Boerne becomes searchable from somewhere else, and it travels with the record by design.

3

The signature outlives the plate. Flock sells matching on vehicle appearance without a readable tag. Cover or swap your plate and the system is built to follow the same car anyway.

4

“Purpose of query” is free text. The field exists and is required. The ACLU of Massachusetts found officers satisfying it with entries like “investigation” and “hehehe.” A required field is not the same as an accountable one.

5

The audit trail is a second dataset. The logs reveal which officer searched whom, under what stated reason, and with what natural-language prompt. Millions of those search records have already been released publicly through police redaction errors.

What could happen if Flock Safety is breached

Five scenarios, none requiring novel technique. Each one has already happened at smaller scale, by insiders, with audit logs running and their names on the query. A breach hands the same file to anyone, with the audit log and the name removed.

01

Rebuild where you sleep, work, worship, and get treated

Filter one plate, sort by time, and a life falls out of the data. The address you return to every night. The building you sit outside 8:45 to 5:15. The parking garage next to the divorce attorney. No hacking of any individual required — just a query and a sort.

This is the capability that makes every other item on this list work. Everything below is what you do once you have it.

PrecedentIn January 2026, police departments’ own redaction failures in public-records releases exposed 2.3M+ plates and tens of millions of search records — movement data spilling in bulk with no intrusion at Flock at all.
Hits hardest
everyoneDV survivorswitnessesundercover officers
02

Hunt a specific person, in near-real time

Domestic abusers do not need a nationwide dataset. They need one: the car she left in. A leaked network tells a stalker which county she drove to, which motel lot the car sat in overnight, and which route she takes now. Shelters and safe houses have street addresses, and cameras sit on the roads into them.

The insider version of this is already documented. The leaked version scales it to anyone with the file and a grudge, with no badge and no log entry.

PrecedentThe Institute for Justice has documented at least 44 incidents of police using ALPR systems to track romantic interests — while logged in under their own names — and calls that “almost certainly an undercount.”
Hits hardest
stalking targetsjudgesjurorsjournalists' sources
03

Automate extortion on inference

Join camera coordinates against a map of clinics, recovery meetings, gay bars, union halls, immigration attorneys, competitors' loading docks. Now sort by target value: executives, clearance holders, elected officials, physicians, clergy. The shakedown writes itself, and unlike the usual sextortion bluff, this one comes with a location log attached.

The inference does not have to be correct to work. It has to be embarrassing enough that the target pays rather than explains.

Why it scalesThe dataset is machine-readable and uniform. One script produces thousands of individualized demands — the same economics that made ransomware an industry.
Hits hardest
executivesclearance holderspublic officials
04

Stalking

The stalking is not hypothetical; it is a documented pattern with names and dispositions. A Sedgwick, Kansas police chief searched an ex-girlfriend’s and her new boyfriend’s plates 228 times over four months. A Louisville officer tracked an ex-partner hundreds of times in two months and was charged with multiple felonies. A Milwaukee officer ran a woman he was dating nearly 180 times. A Braselton, Georgia police chief was arrested and charged for using the system to stalk and harass. A Jerome County, Idaho sheriff searched for his wife’s vehicle hundreds of times. Officers in Greer, South Carolina and Richmond County, Georgia were terminated for the same behavior, and a media review counted 28 officers abusing ALPR systems to track former partners since 2024.

Every one of those took a login and left a trail, and they happened anyway. A leaked copy hands the same capability to any stalker: no badge, no audit log, and no chief to answer to.

PrecedentThe documented cases span Kansas, Georgia, Kentucky, Wisconsin, Idaho, Florida, Texas, and the Carolinas. Every one was caught by the audit log. A leaked dataset has no audit log.
Hits hardest
ex-partnersfellow officersanyone an insider resents
05

Turn it into a physical-world attack tool

Burglary crews want to know which house empties at 7:10 every weekday and stays empty until 6. Cargo thieves want route and timing on the same truck each week. Kidnap-for-ransom needs a pattern, and this is a pattern generator. Meanwhile the camera inventory doubles as a coverage map — the fastest way to plan a route no reader ever sees.

And the system can be attacked in the other direction: get a plate onto a hot list and you have aimed a felony traffic stop at someone. An LAPD Inspector General audit already found 161 of 498 stolen-vehicle alerts were false — 32% — before anyone tampered with anything. LAPD let the contract lapse in July 2026.

PrecedentIn late 2025, researchers found roughly 60–67 Flock cameras livestreaming to the open internet with no authentication, some exposing 30 days of downloadable video.
Hits hardest
homeownersfreight & cash-in-transitanyone framed

Nine scans is a dossier

People underestimate this because they picture a single photo. Below is one fabricated vehicle over 72 hours — nine rows, the volume this network generates for a commuter in three days — and the conclusions any analyst draws from them in under a minute.

Plate 7XKR412 · 72 hours · illustrative data
MON 07:11CAM-4471 · residential collector, Oak & 12th
MON 07:44CAM-2210 · school drop-off loop
MON 08:02CAM-0918 · office park entrance
MON 17:39CAM-0918 · office park exit
MON 18:05CAM-4471 · residential collector, Oak & 12th
TUE 19:22CAM-3355 · church lot, side street
TUE 20:48CAM-3355 · church lot, side street
WED 12:31CAM-7702 · medical campus, garage B
WED 21:14CAM-1180 · motel frontage, Route 9
→

Home address, to within a block — the first and last scan of every weekday.

→

Employer and shift — same gate, 08:02 and 17:39, five days running.

→

A child, and which school — a 33-minute detour through a drop-off loop.

→

A recurring 90-minute Tuesday-night meeting in a church side lot. Bible study, AA, and a support group look identical in this data — which is exactly why the inference is dangerous.

→

A midday medical visit, and the specialty wing it parks nearest.

→

A 9pm stop at a motel on a night the car should have been home.

→

The gaps — the windows when nobody is at that address.

What this looks like in Kendall County

Boerne and Kendall County are not observers of this system. They are nodes in it, and the county's next contract decision is on the Commissioners Court agenda for Tuesday, 25 August 2026. Every figure below comes from the cities' and county's own posted records.

2,215,943 Plate reads logged by Boerne's cameras in a single 30-day window — 360,238 unique plates. Boerne's population is roughly 20,000. Boerne PD staff report (AIS), item 2025-610, 9 Dec 2025
10,856 Cameras Boerne PD can search across 519 shared networks. Sharing runs both directions. Boerne PD staff report (AIS), item 2025-610, 9 Dec 2025
31 Flock cameras operated by the City of Boerne — 12 dedicated plate readers plus 19 existing city cameras running FlockOS. Boerne PD staff report (AIS), 9 Dec 2025 · Res. 2024-R09 · Res. 2025-R90
1,799 Verified signatures on the petition to remove Flock from Boerne and Kendall County, as of 22 Aug 2026. Filed 28 July 2026. Change.org
Kendall County · the record so far
15 AUG 2025Flock agreement term begins — FY2026, $258,305, 80% MVCPA grant-funded
23 SEP 2025Res. R-20250923-1 — accepts grant, approves TxDOT right-of-way use. Passed 5–0
14 OCT 2025County right-of-way license agreement. Passed 3–2 — Wisian and Chapman nay
28 APR 2026Res. R-20260428-1 — new grant application, NTE $200,000. Passed 4–1 — Carpenter nay
25 AUG 2026Items 13, 14, 15 — FY2026 reduced to $212,608.14; FY2027 renewal NTE $198,550; grant $165,458 / county match $33,092
→

The renewal is a live decision, not a settled one. FY2027 runs 1 Oct 2026 – 30 Sept 2027. The vote to authorize it is on this week's agenda.

→

Grant funding is not the same as free. The county still commits a match — roughly $42,500 this year, $33,092 next — and grant-funded programs are the ones counties get stuck renewing when the grant ends.

→

Nobody has published a camera count. Public sources put Kendall County anywhere from 43 to about 60 devices, and no county record states a number. A jurisdiction that cannot say how many cameras it operates cannot audit them.

→

The exposure is not local. A Boerne plate read is searchable through a 519-network, 10,856-camera pool. Every failure documented above — improper federal queries, officer misuse, redaction spills — happened inside pools exactly like it.

→

The question before the court is narrow. Not whether plate readers ever help, but whether Kendall County should keep contributing residents' movement histories to a national pool it does not control, cannot audit, and cannot claw back.

Commissioners Court meets the second and fourth Tuesdays at 9:00 a.m., Commissioners Courtroom, 3rd floor, 201 E. San Antonio Ave., Boerne. To speak, a Public Participation Form must reach the County Judge’s Court Coordinator by 8:00 a.m. the day of the meeting; three-minute limit. No form is required to speak at a posted public hearing.

The point

Every other mass breach can be remediated. You freeze credit, rotate a password, reissue a card. There is no remediation for a record of where you drove for the last eight years — and no way to opt out of a camera on a public road.

The honest framing is not “this network has been breached.” It is that the pieces have already failed independently — cameras open to the internet, hundreds of improper cross-jurisdiction searches in a single city audit, dozens of documented cases of officers tracking romantic interests, a third of hot-list alerts wrong, and millions of plate records spilled by police redaction errors. The full-database scenario is not exotic. It is the same failures, correlated.