What Flock Cameras Track

By DeFlock Signs · Last reviewed July 18, 2026

It is not just license plates

In July 2026, reporters obtained evidence that police had used Flock Safety's cameras to search for a “heavy-set male with a black and white hat,” a “person on skateboard,” and a “person wearing orange vest and construction hat.” Other searches described tattoos, sports clothing, race, and signs of political affiliation. The searches were not hypothetical product demos. They appeared in data reviewed by 404 Media, which found that agencies had used the feature at least hundreds of times, sometimes across hundreds of cameras.

That finding punctures the comforting idea that Flock is only a license-plate company and therefore does not track people. A license plate is already a durable link to a registered owner. But Flock's current system goes much further: it photographs vehicles, extracts searchable visual details, records where and when they appear, reconstructs routes, identifies vehicles that travel together, stores live and archived video, follows pedestrians with pan-tilt-zoom cameras, analyzes event-triggered audio, and can combine sensor data with records held by police and other agencies.

Not every customer buys every product, and not every camera performs every function. That is precisely why the right question is not “What does a Flock camera collect?” It is: What does this Flock deployment collect, derive, import, retain, and make searchable, and who can reach it?

The four layers of a Flock record

Public debate often collapses four different kinds of data into the word “footage.” They should be separated.

1. Raw observations: still photographs, plate crops, live and recorded video, audio clips, and aerial imagery.
2. Machine-generated metadata: plate text, plate state, vehicle type, make, model, color, damage, accessories, time, location, camera identity, and object labels.
3. Inferences and relationships: a reconstructed route, vehicles frequently traveling together, appearances across multiple locations or states, visual similarity matches, and algorithmic judgments about unusual movement.
4. Imported and operational records: hotlists, case numbers, warrant or report information, 911 and dispatch data, jail and records-management data, public records, user identities, search reasons, and audit logs.

The surveillance power comes from joining those layers. A photograph says a car passed one camera. A searchable series of photographs says where it went. A convoy analysis suggests who traveled with it. A records system supplies a name, address, case history, or known associate. Video may show who got out.

What an ALPR sighting contains

A federal system-of-records notice for a Flock deployment at the Presidio describes a surprisingly rich record. Depending on the export format, it can include a full or partial plate reading, the capture time, camera and camera network, latitude and longitude, image filenames, a contextual vehicle image, a close-up plate image, a map of the camera location, plate state, and the vehicle's body type, make, and color. The notice also explains that users can export the data as spreadsheets, image archives, or reports. See the Presidio Trust notice.

Testimony summarized in a federal court record adds technical detail. Flock cameras use infrared imaging and may take multiple high-speed pictures before software chooses the image uploaded to the system. The platform can identify or index the plate, make, model, color, plate type, direction of travel, damage or alterations, bumper stickers, and roof racks. It can photograph motorcycles and bicycles as well as cars. The record states that some camera angles can show that a vehicle has multiple occupants, even if the system cannot identify them. See United States v. Jackson.

This is important for two reasons. First, the database can locate a vehicle even when an officer has only a partial plate or a visual description. Second, “the system does not identify the driver” is not the same as “the system collects no information about people.” Context photographs can incidentally capture occupants, cyclists, pedestrians, clothing, decals, political bumper stickers, disability placards, work equipment, and other clues. A Bartlesville police FAQ, for example, acknowledges that a person can appear in an ALPR still image, even though the agency said its then-current system could not search by person. The July 2026 FreeForm records show that this limitation no longer describes the whole Flock platform.

A practical inventory

From sightings to movement profiles

Flock representatives often distinguish a point-in-time camera observation from continuous GPS tracking. Technically, that distinction is real. A fixed ALPR does not attach itself to a car. But it can still produce a movement history when many point observations are linked.

In a June 2026 investigation, InvestigateTV reviewed Flock training videos in which instructors and officers described tracking a vehicle “from location to location to location,” tracking a suspect's movements, and following a suspect across state lines. The reporters also tested a Condor camera in the field and watched it physically follow a pedestrian.

The platform's analytical tools deepen the movement record. Public audit records cited in a 2026 San Jose lawsuit list FreeForm search, visual search, and convoy analysis. Convoy analysis identifies vehicles repeatedly seen near a target vehicle. Multi-location and multi-state tools find vehicles that appear in particular combinations of places. The ACLU's analysis warns that these features do more than retrieve a known vehicle: they can generate suspicion by flagging movement patterns and associations.

That means the platform can infer relationships that no camera directly observed. Two cars repeatedly traveling together may belong to accomplices, coworkers, family members, neighbors with similar commutes, or strangers caught in the same traffic flow. The database supplies correlation; the officer or algorithm supplies meaning.

The consequences are not abstract. An Associated Press investigation found that Border Patrol used license-plate networks and predictive analysis to flag supposedly suspicious travel patterns, leading to stops and searches. Documents showed that CBP or Border Patrol had, for a time, access through Flock to at least 1,600 readers in 22 states. The AP also described an official boasting that the system could reveal a vehicle's path through Dallas, Little Rock, Atlanta, and south Texas.

Flock cameras can collect and track people directly

The clearest contradiction to “vehicles, not people” is Flock's Condor video system. Condor is a pan-tilt-zoom camera, not a conventional plate reader. InvestigateTV watched one detect a reporter, pan and tilt as he walked, and keep him in the frame. A training video reviewed by the outlet said the camera can acquire and track a person or group after detecting human movement.

An earlier investigation by 404 Media and security researcher Benn Jordan found more than 60, and eventually roughly 70 Condor devices streaming without passwords. Investigators watched cameras follow people on trails, in parking lots, and near homes, and were able to retrieve archived video. The reporting showed that the video could reveal faces and very fine detail; one camera captured what a person was watching on a phone. InvestigateTV's account is available here, and an independent summary of the original 404 Media investigation is available here.

FreeForm makes that footage searchable with ordinary language. According to 404 Media's July 16, 2026 report, officers searched for people by build, hats, vests, sports shirts, tattoos, activity, race, and political signals. This is people search even if the software never assigns a legal name or constructs a biometric face template.

That distinction matters. People tracking is not synonymous with facial recognition. I found no reliable independent evidence that Flock's own platform currently performs biometric face matching. But the absence of proven facial recognition does not erase what is documented: high-resolution faces are captured; cameras automatically follow bodies; and AI search can retrieve people based on visible traits. The public portion of 404 Media's evidence does not establish whether traits such as race are stored as permanent labels or evaluated when a user submits a search. Either design can still return footage of people matching the description.

Flock also listens

Flock's Raven product adds environmental audio. It began as gunshot detection and was later expanded to detect “human distress,” including screaming, according to The Record and the Electronic Frontier Foundation.

The evidence supports a careful description. Raven sensors must analyze ambient sound in order to decide whether an event occurred, but that does not prove they upload a continuous cloud recording. A Wichita Police Department policy states that an activated sensor creates a five-second FLAC audio file for each alert and that additional detections produce additional files. It says the platform retains those clips for 30 days unless police download them to Evidence.com. For an event, Flock can also provide associated audio files, a map, GPS locations, and a list of Raven devices.

So the defensible conclusion is not “Flock records every conversation.” It is that Flock operates microphones that continuously evaluate public sound and retain short clips when the system classifies an event, including, now, a category involving the human voice. Accuracy, false-trigger rates, and the scale of distress-detection deployment remain poorly documented by independent sources.

Drones and integrated police data expand the view

Flock's Drone as First Responder system turns the platform into an aerial sensor. Public procurement records describe remote piloting, live visual feeds, thermal imaging, night vision, powerful zoom, laser rangefinding, flight logging, mission reporting, and integration with 911 or computer-aided dispatch. See the records from Madera and Greenfield, as well as Axios's account of the Aerodome integration.

The resulting data can include live or recorded aerial imagery, heat signatures, camera zoom and orientation, drone position and telemetry, launch time, flight path, mission details, and the call for service that caused deployment. Exactly what is recorded and how long it is kept depends on the agency's configuration and evidence policy; there is no responsible basis for applying a single retention claim to every drone program.

The Flock platform can also ingest information that its own roadside sensors never observed. Depending on the customer's integrations, public procurement records show it can display 911 and dispatch events, their location, priority and event type, live patrol locations, and data from records-management and jail systems. This creates a feedback loop: a 911 call can launch a drone; an ALPR hit can cue officers or another sensor; video can be searched for a person; and the resulting evidence can be attached to a case.

Nova is designed to link a plate to a person

Nova, Flock's investigative platform, makes the “vehicles, not people” distinction even harder to sustain. Internal meeting audio reviewed by 404 Media described an ambition to “jump from LPR to person” and connect that person to others through relationships such as marriage or alleged gang affiliation. The design contemplated public records, police records-management and dispatch data, commercial data sources, and breached information that could connect a plate with an email address, phone number, mailing address, or identity.

The hacked-data detail requires precision. After the internal plan became public and employees objected, Flock announced that Nova would not supply data purchased from breaches or the dark web. Independent evidence establishes that the company explored the idea; it does not establish that hacked data is a current production input. Nova nevertheless remains a people-lookup and linkage tool through public records, open-source intelligence, agency records, and other customer-selected sources. In April 2026, Axios reported that San Diego police had quietly signed a Nova pilot; the department said its particular pilot would not pull in ALPR data, illustrating how configurations can differ.

Flock collects data about the watchers, too

Every search creates another sensitive record. An audit analysis by the University of Washington Center for Human Rights describes fields including the user's identity and agency, search date, time range, number of networks and devices searched, full or partial plate, stated reason, case number, and filters such as vehicle make or state. Event logs may also record hotlist creation, permission changes, password resets, and network-sharing settings.

Those logs are essential for accountability, but they also reveal investigative targets and police interests. In June 2026, 404 Media found that Flock pages had exposed some search reasons and, in some cases, the searched license plates to DuckDuckGo and Bing. A security control had become a second leakable database.

The logs also show how weak a mandatory “reason” field can be. The Houston Chronicle analyzed about 470,000 searches and found that two-thirds used vague variants of “investigation” or “suspect.” Other entries were blank, gibberish, “random,” “donut,” or similarly meaningless. Houston officers could search as many as 88,000 public and private cameras nationwide. The interface recorded a reason; it did not ensure the reason was adequate.

Watchlists are another database, not a momentary alert

Flock systems compare plate sightings against national, state, and local lists and allow agencies to create their own custom hotlists. Agency policies show that a custom entry can include a plate or vehicle description, the reason for interest, report or case number, warrant number, requesting officer, notes, expiration date, and the personnel subscribed to alerts. An alert then adds a detection time, camera and location, matched image, notification recipients, and response history.

The Oconomowoc Police Department's policy lays out many of these fields. A Louisville Metro Police policy requires notes and an expiration date for manual hotlist entries and acknowledges that stale or unverified alerts can lead to improper stops.

The list itself can be more sensitive than the plate image. It records that an agency considers a vehicle connected to a person, warrant, missing-person case, or investigation, an assertion that can be wrong, outdated, or based on an association rather than wrongdoing by the current driver.

“Deleted after 30 days” is an incomplete answer

Thirty days is a common hosted-retention period for raw ALPR images in the court records and agency policies reviewed for this article. It should not be treated as a universal or complete deletion rule.

There are several ways data outlives that window:

• Officers can download images, spreadsheets, maps, or PDF reports before automatic deletion.
• Data moved into a case file, records-management system, or evidence platform follows a different retention schedule.
• Louisville permits downloaded ALPR records to remain for 90 days and longer for an ongoing investigation, evidentiary use, statistics, training, or system evaluation.
• Wichita's five-second Raven clips move from a 30-day Flock window into Evidence.com when preserved.
• Condor investigators were able to retrieve approximately 30 days of archived video from exposed devices, but other video settings may differ.
• Search and audit records can persist after the underlying image disappears.

The Houston Chronicle found that Flock sharing arrangements allowed other agencies to maintain search logs indefinitely. Those logs still contained information about vehicles Houston police had sought, even when the original plate images were subject to shorter retention. Deleting the observation does not necessarily delete the fact that someone searched for it, exported it, received an alert, or attached it to a case.

The network makes local collection national

A city may install 20 cameras and still expose residents to a far larger search system. Houston police had access to as many as 88,000 cameras. In the abortion-related investigation that triggered an Illinois inquiry, a Texas sheriff's request went to a nationwide network of roughly 83,000 cameras, according to the Associated Press. Illinois' Secretary of State later reported that a separate audit found Flock had allowed Customs and Border Protection to access Illinois plate data in violation of state law. See the state's August 2025 audit announcement.

“The customer owns the data” therefore answers a contract question, not a surveillance question. Flock still hosts or processes the records, and customer-controlled sharing can place them within reach of thousands of other users. Ownership does not describe how many copies exist, which agencies can query them, what an officer can export, or how long derivative records survive.

What communities should demand to know

The best public inventory is specific to the deployment. Officials should be required to answer, in writing:

1. Which Flock devices and software modules are active: ALPR, Condor video, FreeForm, convoy or multi-state analysis, Raven audio, drones, Nova, 911/CAD, or other integrations?
2. What exact raw fields, AI labels, and derived relationships does each module create?
3. Can the system search people by clothing, tattoos, race, political symbols, activity, or uploaded images?
4. Does any camera automatically pan, tilt, zoom, or follow a person? Is video archived, and for how long?
5. Which sounds trigger retained audio, how long are clips, and where are preserved clips copied?
6. Which public, commercial, police, jail, dispatch, or open-source databases feed Nova or the broader platform?
7. Which hotlists are enabled, who may add entries, what fields are required, and when do entries expire?
8. Which agencies, private-camera networks, state systems, and federal users can search local data—directly or through an intermediary?
9. Which records are deleted automatically, which can be exported, and what are the retention schedules for exports, alerts, cases, and audit logs?
10. Who independently audits search reasons, permissions, sharing changes, false matches, and system accuracy—and are the detailed results public?

Flock is not merely collecting plates. It is building structured, searchable context around movements, associations, sounds, incidents, and people. The most important data may be a license plate, a tattoo, a five-second scream, a convoy match, a 911 location, or the identity of the officer who searched. What turns those fragments into surveillance is the platform that links them—and the network that lets a local observation travel far beyond the camera that made it.

Source note

This article intentionally does not rely on Flock Safety's own website, FAQs, privacy statements, marketing pages, or product manuals as evidence. It uses independent reporting, court and government records, public agency policies, public-records research, and independent technical investigation. When a government record repeats a vendor capability, the article treats it as evidence of what that agency procured or represented to the public, not proof that every Flock deployment is configured identically.

Research current through July 18, 2026.