Your daily drive feeds the dragnet.

Automated license plate readers, or ALPRs, collect passing traffic without first singling out a suspect. A plate, image, time and location become a record that can be searched later. Linked cameras and agency sharing can turn scattered sightings into a history of someone’s movements. You do not have to be on a wanted list for your trip to enter the database.

Plate textABC · 1234

Seen7:42 PM

Camera014 · Westbound

VehicleBlue SUV · Roof rack

Now a searchable location record

Illustrative record. The plate, time and camera location are fictional.

Collect broadly. Search later.

Collection comes first. Cameras record ordinary traffic within view. A car does not have to be stolen, wanted or connected to a crime to be recorded.

A history accumulates. Searchable locations can reveal routines and visits to sensitive places. Other records can connect a plate to a name. Read EFF’s explanation.

The reach expands. Sharing can put local sightings within another agency’s reach. Exports can enter separate systems with their own analysis and retention. See the proposed next layer.

What happens when one vehicle passes

This is the common data path. Exact fields, features, and retention differ by vendor and agency.

1
Capture
A roadside, parking-lot, or patrol-car camera takes images of a passing or parked vehicle and the surrounding scene. People in view can appear in those images.
2
Translate
Software uses optical character recognition to turn the plate image into searchable letters and numbers.
3
Build a record
The system selects useful frames and pairs the read with the date, time, camera location, and scene or vehicle imagery. Some systems add color, make, body type or direction. Separate video products may support people searches; features vary by camera and configuration.
4
Compare and alert
The software can compare the read with a hot list and notify authorized users of a possible match. Reads that trigger no alert can still be retained for later searches.
5
Search and connect
Stored reads can be searched later. Sharing and integrations may expand that search beyond the camera owner's own system.

Recovered-device evidence / one Flock camera

The plate reader also photographed and detected people.

A September 16 joint investigation by 404 Media and WIRED examined files copied after hackers physically removed one Flock camera. The device took different exposures of the plate and the wider roadway scene. A typical passing vehicle generated about 28 images, and some generated more than 100.

The software on that camera explicitly looked for people, vehicles, license plates, and bicycles. Reporters found people in 11 recovered video clips, all riding motorcycles. Camera placement matters: this unit was pointed down at a roadway, where pedestrians were unlikely to pass through the frame.

Detection is not identification. The investigation found no evidence that facial recognition was enabled or actively used. But “no facial recognition” does not mean the camera captures only a plate or that people are absent from its images.

Read the 404 Media investigation Read the WIRED version

1.6 million images: Generated from roughly 50,200 vehicle events in about 21 recovered days on one camera.
27,321 short clips: One- to two-second, 1,024 by 768 video files without audio were recovered separately from higher-resolution image bursts.
People explicitly detected: The device software recorded where a detected person appeared and its confidence in that detection.
One camera, not every camera: The findings establish what this recovered device did. They do not prove every deployed model has the same configuration or exposure.

Faces, filed as “objects.”

404 Media’s September 29 investigation describes VIDIZMO’s “Object Library,” where uploaded faces become entries for facial recognition. The company pitched police a platform combining Flock and Axon material with searches by face, vehicle or object.

Enroll a face.: VIDIZMO’s documentation tells users to upload faces into its library and add people to watchlists.
Put a name on the image.: The company says enabled facial recognition can match a face in a camera feed to an enrolled person.
Trigger tracking and recording.: The documented workflow can generate watchlist alerts and capture a clip around a recognition event.

The Flock connection was a proposal. VIDIZMO’s CEO told 404 Media the connector had not been built and it had not run facial recognition on Flock footage. Its integration page still describes a connector built on request.

The risk is clear: collected footage can become the input for another company’s surveillance tools. A camera vendor’s promises alone do not settle what happens to exported material.

Read 404 Media’s investigation · Read VIDIZMO’s proposed integration

What the system may know or do

“Available” does not mean “enabled here.” Check the agency’s contract, settings, policy, law, and audit records.

A basic read commonly records

images of the plate, vehicle, and surrounding scene, which can include people
the letters and numbers the software thinks it saw
date and time
camera location
sometimes plate state and direction of travel

Some products also offer

make, model, color, body type, and visible vehicle features
full, partial, temporary, or even missing-plate searches
plain-language searches for vehicles, objects, or people using visible descriptions
person and object detection, cropping, and image-comparison tools
live alerts, maps, route views, and repeated-sighting analysis
cross-agency sharing and searches across partner cameras
connections to video, dispatch, records, and real-time crime-center tools

One plate querycan return many past sightings

One hot-list entrycan trigger a future alert

One local recordcan become available to partners

Many sightingscan expose routes and routines

The system records a vehicle. People can infer a life.

A single read is a sighting. Repeated reads can show when a vehicle passed particular cameras and where it was seen. Over time, a user may infer regular routes, likely home or work locations, visits to sensitive places, or vehicles that repeatedly appear nearby.

Those conclusions are inferences, not facts automatically proved by the software. A plate can be connected to a registered owner through another database, but that still does not establish who was driving on a particular trip.

Directly recorded: Scene and vehicle images, translated plate, date, time, camera location, and available vehicle, person, or object details.
Computed or inferred: Possible route, repeated presence, common locations, or association with other vehicles.
Linked from elsewhere: Registered owner or investigative information obtained from a separate government or commercial database.

A match is not proof

Optical character recognition can misread a plate. Hot-list information can be wrong, stale, or attached to a similar plate. An image may show a rider, driver, passenger, pedestrian, or bystander, but that does not establish identity. Flock’s own policy says users should confirm the computer translation before acting on an alert or search.

An alert is an investigative lead, not a finding of guilt.
A plate sighting does not show why the vehicle was there.
A registered owner is not necessarily the person driving.
A system’s confidence score is not independent verification.

Read the vendor policy language

Evidence and source limits

Warnings from reporting, testimony and court records

Updated October 16, 2026

Five witnesses described risks from vehicle-location databases, wrongful investigations, weak oversight and security flaws. Their testimony is evidence of what they told Congress. It is not a committee finding, a court ruling or a new law.

September 23 hearing and five written statements · Watch the hearing clips

Witness testimony

A vehicle sighting is a lead, not proof of guilt

Lindsey Isaacs told senators that a photograph of her actual vehicle near a fatal crash helped focus an investigation on her despite contrary evidence. She described 13 days in jail before the State declined to file an Information. Her account does not establish that the camera misread her plate. Investigators still need to check the vehicle, driver, time, location and other evidence.

Lindsey Isaacs — September 23, 2026

Legal institute testimony and advocacy

Repeated sightings can reveal more than a plate

Institute for Justice counsel Alasdair Whitney warned that accumulated sightings can expose routines, associations and sensitive destinations. These are inferences drawn from location records; a camera does not directly know why a person visited a place. Wider sharing and historical searches can increase what investigators learn.

Alasdair Whitney — September 23, 2026

Official summary of hearing testimony

No facial recognition does not mean no people appear

In an exchange summarized by Senator Durbin's office, Pinal County Sheriff Ross Teeple said he recognized people in images his department accessed. That describes a person recognizing someone in a photograph. It does not establish that the system used automated facial recognition.

Senate Judiciary minority office — September 23, 2026

Vendor proposal and original investigative reporting; checked September 29, 2026

Outside software can expand what camera data is used for

VIDIZMO markets a proposed FlockOS connector alongside products offering watchlist facial recognition and behavioral analysis. Its integration page says the connector is not prebuilt; live analysis also depends on reachable camera streams. A camera vendor’s stated limits do not describe every outside tool that could process exported material. Ask which exports, integrations and secondary uses an agency permits. These sources do not establish deployment in Prescott Valley or Johnson City.

VIDIZMO proposed integration · Additional source

Vendor rollout announcement and original CEO interview; checked October 16, 2026

A search log is not an audit

Flock announced mandatory case codes and Audit Assistance for law-enforcement customers by year-end 2026. An announcement does not prove local controls are enabled; a logged reason is not a completed audit. On October 14, Flock's CEO told TechCrunch the company had not terminated a police-agency customer over product misuse, while describing circumstances in which it might. That is his account, not an independently verified contract audit. Ask for actual settings, findings and corrective actions.

Flock safeguards announcement and CEO interview · Additional source · CEO interview

Security researcher testimony and vendor response

Security flaws and fixes both need scrutiny

Researcher Benn Jordan described device and video-interface weaknesses to the Senate. Flock separately says a 2026 Bishop Fox test found 36 issues and that 24 were fixed and retested, including all critical and high findings. The full reports were not independently reviewed here. A security test also does not establish that agency searches were lawful.

Jordan testimony; Flock disclosure September 22, 2026 · Additional source

Legal institute testimony and advocacy

Surveillance can affect lawful activity

ACLU counsel Chad Marlow warned that a searchable record of movements can discourage people from attending protests, worship services or other lawful activities. This is a civil-liberties warning and policy argument, not a judicial finding that every camera has caused those harms.

Chad Marlow — September 23, 2026

Why full audit logs matter: officer misuse cases and agency responses

What the equipment collects depends on the product

What an ALPR directly records

Flock describes plate and vehicle images, visible vehicle characteristics, a timestamp and camera location. Other records may then connect a plate to an owner. An owner is not necessarily the driver.

Vendor description; local configuration varies

Video cameras are a separate capability

Flock's video products provide live and recorded video, zoom controls and people-detection alerts. Video and ALPR cameras can share a platform or pole. The presence of a plate reader does not establish that a town also bought these video features.

Vendor product documentation

People search can mean clothing or visible features

Flock says FreeForm can search enabled video for visible characteristics such as clothing or accessories. Its documentation distinguishes this from biometric identification and says People FreeForm is not offered on LPR-camera images. A statement that Flock never searches for people is therefore too broad; the camera type and enabled features matter.

Vendor product documentation; not proof of local deployment · Product scope

Audio detection uses separate sensors

Flock describes short sound windows processed on the device and clips retained for qualifying events. It says the system does not transcribe speech or identify speakers, although an event clip can incidentally contain words. Verify whether a local agency has audio sensors before attributing this capability to it.

Vendor description

Check retention after data leaves the local account

Flock’s 7 day default applies only to new law enforcement customers. Existing and private accounts can differ. 404 Media reports that some agreements send local ALPR data to HIDTA regional systems that can share with the DEA’s national program. Verify local settings, exports, regional or federal copies, and what is deleted when a contract ends. This does not establish that every local reader feeds HIDTA.

Vendor policy and records-based federal-sharing investigation · Product scope

What courts have and have not decided

A ruling about one kind of location record does not automatically decide every ALPR case. The court, jurisdiction, facts and procedural stage matter.

2026-09-28 · Jury verdict reported by plaintiff’s counsel; not appellate precedent

Schott: surveillance does not justify an unlawful stop

According to the Institute for Justice, a federal jury found Bexar County liable for a policy that caused an unconstitutional stop, detention and search of Alek Schott. The verdict concerns that program, not all ALPR use. IJ said it would seek an injunction; final judgment and any later relief remain under verification.

Read IJ’s verdict report

2026-06-29 · Supreme Court holding — geofence/location data, not ALPR

Chatrie: private location data can remain protected

The Supreme Court held that obtaining the Google Location History at issue was a Fourth Amendment search and returned the case for further proceedings. It did not decide the validity of the warrant or rule directly on ALPR systems.

Read the decision · Read IAPP’s analysis

2018-06-22 · Supreme Court holding — historical cell-site records

Carpenter: historical tracking raises different concerns

The Supreme Court held that obtaining historical cell-site location information was a search generally requiring a warrant. The opinion concerns phone-location records and is not a blanket ruling against plate readers.

Read the decision

2021-06-24 · Fourth Circuit en banc holding — aerial tracking

Baltimore aerial surveillance: the scale matters

The Fourth Circuit treated access to Baltimore's retrospective aerial movement database as a search and reversed the denial of preliminary relief. The decision concerned that aerial program, not every ALPR installation.

Read the decision

2026-10-02; 2026-10-01; 2026-03-17 · Conflicting trial-court rulings and Fifth Circuit precedent

ALPR cases turn on the search, record and jurisdiction

An Oklahoma federal trial court suppressed evidence after finding the month-long Flock search in United States v. Kyle unconstitutional. The Fifth Circuit upheld the narrower query in Porter. A California state trial judge dismissed San Jose organizations’ state-constitutional challenge on October 2; appeal was only under consideration when reported. None of those decisions settles every ALPR use, and only Porter is binding appellate precedent.

Read the Kyle suppression order · Read Porter for comparison · Read the San Jose dismissal report

2024-04-15; 2025-03-20 · Federal trial-court dismissal and postjudgment order

An accurate plate read can still lead to a stale alert

Smith v. Flock Safety describes an accurate plate match to an outdated police hot-list entry. The Ohio trial court dismissed the claims against Flock and later denied reconsideration. This vendor-liability ruling is not a finding that every alert is reliable or every ALPR search constitutional. Officers still need to verify the alert and its current basis.

Read the decision · Related decision

Court records, appeals and counsel

Three things people often mix up

ALPR is not automatically facial recognition.: Flock says its LPR product does not use facial recognition. Detecting a person, searching by visible description and identifying someone by face are different functions. Separate video and outside analysis tools can expand what the footage is used for. See VIDIZMO’s proposed addition and the records needed to establish local use.
ALPR is not exactly a GPS tracker.: It does not usually transmit continuously from the car. It creates sightings wherever participating cameras see the vehicle. Dense coverage, broad sharing, and longer retention can still produce a detailed movement history.
ALPR is not necessarily a ticket camera.: Some plate-reading systems support tolling, parking, or traffic enforcement. The systems followed by this site are primarily used for surveillance, alerts, investigations, and data sharing.

What Flock’s patents describe

A patent records systems the company sought the legal right to protect. It is evidence of contemplated technical architecture, but it does not prove that every described feature was built, sold, enabled, or used by a particular agency.

US 11,030,892 B1: the camera and remote analyzer: The patent describes neural networks that filter captured images, send selected images to a remote analyzer, compare plate information with other databases, generate alerts or reports, and receive updated neural-network weights from the remote system. “Neural network” means image-analysis software.
US 11,416,545 B1: DSOQ: DSOQ means Dynamic Surveillance and Object-Based Query. The patent describes combining video from distributed sources, classifying objects and characteristics, searching historical footage, producing timelines, and generating alarms. Its examples include classifying people by attributes that include race and searching separate home-security footage using matching facial-recognition data points.
The records needed to prove local use: A patent cannot establish what Prescott Valley or another customer actually operates. That requires contracts, product schedules, system architecture, enabled-feature lists, camera and gateway inventories, integrations, administrator settings, audit logs, exports, retention records, and vendor-access records.

Read US 11,030,892 B1 Read US 11,416,545 B1

The software layer, in the vendors’ own words

These examples describe advertised or documented capabilities. They are not endorsements and do not prove a feature is active in any particular jurisdiction.

SystemDocumented capabilitiesSource

Flock LPR and FreeFormPlate and vehicle-attribute searches, hot-list alerts, cross-jurisdiction sightings, network sharing, and, in FreeForm, plain-language searches across enabled LPR and video evidence. 404 Media reports that police have used FreeForm to search for people by clothing, build, tattoos, activity, race, and other visible descriptions.LPR FreeForm

Rekor ScoutPlate, make, model, color, direction, live map, configurable retention, and searches using full, partial, or missing plate information.Product page

Motorola VehicleManagerHot-list alerts, batch plate searches, mapping, agency data sharing, retention controls, natural-language assistance, image comparison, and case management.Release notes

The switches that determine how far it reaches

Hardware count matters. These settings and policies often matter just as much.

Retention: How long ordinary reads remain searchable before deletion.
Sharing: Which agencies, private customers, regional systems, or federal users can request or receive access.
Search rules: Whether a user must enter a case number, reason, crime type, or supervisor approval.
Hot lists: Who may add a vehicle, which outside lists are used, and when entries expire.
Audit logs: What is recorded about each search, alert, export, or account—and who actually reviews it.
Exports: Whether images or result files can be copied into another case system and retained separately.
Integrations: Whether ALPR data appears beside dispatch, video, records, real-time crime-center, or other surveillance feeds.
Vendor access: What support staff or privileged administrators can access, and under what conditions.

Ask about the system, not just the cameras

The clearest public record usually comes from the contract, data-sharing settings, audit logs, retention rules, hot lists, exports, integrations, and actual search history.

Build a records search

How many cameras are active, and who owns them?
What fields does each read contain?
Does the device save scene images or short clips, and does it detect or crop people?
Can users search by appearance, or enroll faces in object libraries and watchlists? Who controls those lists?
How long are ordinary reads kept?
Which outside agencies or private systems can search the data?
Must users enter a case number and a meaningful reason?
Who reviews audit logs, how often, and what happens after misuse?
Can records be exported or retained in another system?
What other video, dispatch, federal, or intelligence systems are connected?

Sources behind this guide

Updated October 1, 2026. Vendor pages and patent records document claimed or contemplated capabilities. Technical investigations, government reviews and legal institute sources provide independent context.

Joint technical investigation

404 Media and WIRED: recovered Flock camera files

A joint analysis of files copied from one physically removed camera found rapid image bursts, short video clips, wider-scene exposures, explicit person detection, and an encryption key that unlocked some stored media. The reporters found no evidence that facial recognition was enabled or actively used.

Investigative reporting