Flock is more than license-plate readers.

Last reviewed: September 19, 2026

Flock is best known for automated license-plate readers. That is no longer the whole product line.

The company also sells:

acoustic sensors formerly branded Raven and now called Flock Gunshot Detection;
live and recorded video cameras formerly branded Condor;
software that brings audio alerts, video, license-plate observations and other evidence into the same platform;
and additional products such as mobile camera trailers and drones.

Product names change. The important questions are what a device collects, what software can do with it, how long information remains available, who can access it and what other systems it can be connected to.

The short version

Raven does continuously analyze sound, but available evidence does not support saying it records every conversation. Flock says the current system analyzes short rolling windows on the device, discards ordinary audio and saves short clips when it detects a qualifying event. A limited pilot also detected screams or other sounds of human distress, but Flock says it ended that feature in 2026.

Condor can automatically detect, follow and zoom toward people, but that is not the same claim as facial recognition. Flock materials describe live and recorded video, person-detection alerts, remote pan/tilt/zoom, a dual-lens model with a 32× zoom view, cloud search and integration with other Flock products. Flock says its video products do not use facial recognition.

Those distinctions matter. Our case should be accurate enough to survive scrutiny.

Capability matters even when present use is unknown

We distinguish what the technology can do from what a municipality is currently documented as using. That protects credibility, but it does not make unused or undisclosed capability irrelevant.

Software features, subscriptions, integrations, alert categories and sharing permissions can change without replacing a roadside device. Unless a municipality publishes current entitlements, configuration records, change logs and meaningful audits, residents may be unable to verify when a capability was enabled, disabled or used.

So both statements can be true:

we should not allege that a local agency uses a feature without evidence;
and residents should evaluate the full technical capacity that the installed system could support later.

The right question is not only “What do you use today?” It is also “What can this system be made to do, who can authorize the change, and how would the public know?”

Raven: what the acoustic sensors actually do

Flock says its gunshot-detection sensors continuously analyze sound in rolling windows of about three seconds. According to the company, audio that does not match a qualifying event is not retained or transmitted. When the system detects an event, it saves an encrypted short clip, estimates the location and sends an alert.

The current event list described by Flock includes gunshots, fireworks, vehicle crashes and sounds associated with sideshows such as sustained tire screeching. In a 2025 security notice, Flock said a detected event produced a three-second audio clip stored in the cloud for 30 days. Retention can change by product, contract, customer setting or later policy, so local records still matter.

Flock: How Gunshot Detection Works →

Flock: Gunshot Detection and LPR Security Notice →

Did Raven listen for human voices?

For a limited pilot, yes—but only in a specific sense.

Flock says it tested detection of sounds of human distress, including screaming. The company says the feature did not understand words, identify speakers, transcribe speech or provide a live listening feed. In 2026, Flock said it removed the pilot feature rather than releasing it broadly.

That history is relevant because it shows that the categories an acoustic system is designed to detect can expand through software. It does not establish that current Raven sensors secretly transcribe ordinary conversations.

Raven privacy and safety concerns

The first concern is not a fictional live wiretap. It is the real placement of continuously operating acoustic classifiers in public neighborhoods.

Questions a community should ask include:

What sounds are the devices currently configured to detect?
Who can change those categories, and does a change require a public vote?
Exactly how much audio is saved before and after a detected event?
Can speech be incidentally captured in an event clip?
How long are clips, locations, alerts and related metadata retained?
Who can play, download, export or share them?
Are diagnostic samples collected, and under what written limits?
What are the local false-positive and false-negative rates?
What happens when fireworks, construction, a car backfire or another sound is classified as gunfire?
Does an alert automatically surface nearby vehicles from the ALPR network?

Flock acknowledges that an event clip can contain incidental nearby words and that a confirmed sound event can surface vehicles seen entering or leaving the area. That combination can turn one acoustic classification into a much broader inquiry involving people who happened to drive nearby.

A false alert is not merely a technical statistic. Depending on local response practices, it may bring armed officers into a neighborhood based on an automated classification. Human review helps, but it does not eliminate the need for public accuracy data, response rules and after-action auditing.

Condor: live video, recorded video and zoom

Flock’s current video-camera page describes:

live and recorded video;
fixed cameras with digital zoom;
remotely controlled pan/tilt/zoom cameras;
people- and vehicle-detection alerts;
cloud search and sharing;
search based on visible vehicle characteristics;
integration with license-plate readers, acoustic sensors and drones;
and a dual-lens PTZ model with a 90-degree overview and a 32× zoom view streaming at the same time.

Flock: Video Cameras →

Flock’s August 2024 announcement for Guardian Mode is more specific. It says Condor can automatically detect and track people and vehicles, follow their movement and zoom in to capture identifying details without an operator continuously steering the camera. Flock describes the resulting captures as fixed points in time rather than continuous tracking. It is still automated AI-assisted people tracking inside a camera’s field of view, even though it is not facial recognition.

Flock: Guardian Mode automatically tracks and zooms →

An official Berkeley, California surveillance-acquisition report provides a useful independent government description of a proposed Condor deployment. It describes high-definition video, remote pan/tilt/zoom, people-detection alerts, vehicle tracking across feeds, nighttime imagery, cloud storage and use of the same FlockOS platform as the city’s ALPR system. Berkeley’s proposed rules prohibited facial recognition and audio for that deployment.

City of Berkeley: Condor acquisition report, pages 45–54 →

Does Condor zoom in on faces?

Yes. It can zoom in on a scene and may capture an identifiable face when distance, angle, lighting and image quality allow. Flock promotes both remote PTZ control and Guardian Mode’s automated tracking and zoom.

That is different from saying Condor identifies a person by automatically comparing a face against a biometric database.

We have not found support for claiming that the current product performs facial recognition, and Flock says it does not. Berkeley separately prohibited facial recognition in its proposed use policy.

No facial recognition does not mean no people surveillance.

An operator may follow a person with PTZ controls, and Guardian Mode can automatically follow movement in view. Software may alert when a person appears. Recorded video may show clothing, companions, behavior, direction of travel, a vehicle, a phone screen or a destination. Enabled search tools may find people by visible non-biometric characteristics. Those observations can be revealing even when nobody converts a face into a biometric template.

Documented Condor exposure put live and archived people video on the open internet

On December 22, 2025, 404 Media reported that researchers found at least 60 Condor camera interfaces exposed to the public internet without a password or login. The exposed pages provided live views and roughly 30 days of archived video; reporting also described access to settings, logs or other administrative functions on some interfaces.

The footage made the risk concrete. It included children on a playground, a woman walking a dog on a path, a man in a parking lot and a rollerblader who appeared on more than one camera. Cameras were observed automatically zooming toward or following people. Reporter Jason Koebler visited one of the camera locations and watched himself appear in the live system while colleagues viewed it remotely, independently verifying that the exposure was real.

Benn Jordan demonstrated how imagery from an exposed camera could be combined with facial-search and other open-source information to identify people and learn personal details. That does not mean Condor itself performed facial recognition. It shows how high-resolution surveillance imagery can become identifying when somebody combines it with tools and information available elsewhere.

The researchers found the interfaces through Shodan, a public search service for internet-connected devices. This was therefore publicly discoverable exposure, but it is more precise to say “findable through a public internet-device search” than “a normal Google search.”

Flock told reporters that the problem resulted from a limited configuration used during carrier testing, affected a small number of cameras and was corrected, with additional safeguards deployed. We have not found evidence that every Condor camera was exposed or that Flock’s central cloud platform was breached.

No reported abduction resulted from this incident. The documented access nevertheless created a real child-safety and stalking risk: a stranger viewing a live or archived feed could potentially learn when a child or adult appeared alone, where they traveled and whether a routine repeated. That is a risk the exposed access made possible, not an outcome we claim occurred.

404 Media: exposed Condor feeds, archives and automatic tracking →

WPTV: Flock confirms and responds to the camera-feed exposure →

Watch Benn Jordan’s investigation →

Being watched can change lawful behavior

Researchers often use the term Hawthorne effect for behavior changes that occur when people know they are being observed. A systematic review found evidence of an effect in some studies, but also warned that definitions, methods and effect sizes vary. A newer review focused on video recording found that some participants reported changing their behavior, while emphasizing substantial uncertainty and study limitations.

For public surveillance, chilling effect is often the clearer phrase. People may avoid a lawful protest, clinic, place of worship, support meeting or private conversation when they believe an identifiable record could be watched, searched or reviewed later. The evidence does not justify saying every camera always changes every person’s behavior. It does justify asking whether pervasive observation imposes a social cost even when no one is accused of a crime.

Systematic review of the Hawthorne effect →

Systematic review of behavior change around video recording →

The integrated platform is the larger issue

Flock markets these tools as a connected ecosystem. Its own materials describe integrating video with license-plate readers, gunshot detection and drones. Earlier product materials describe FlockOS as a cloud-based counterpart to a real-time crime center.

Flock: Raven, Condor and FlockOS product announcement →

Consider what an authorized user could potentially assemble when products are connected:

An acoustic sensor creates an alert at a place and time.
Nearby ALPR observations surface vehicles in the area.
Live or recorded video provides people, clothing, companions and direction of travel.
Other cameras or shared data provide later sightings.
Search and alert tools make the resulting evidence easier to revisit, distribute and act on.

That is a capability scenario, not proof that every Metro Detroit agency has purchased, enabled or used every step. The correct local question is what is actually installed, licensed, configured, shared and used here.

It is also fair to ask what the installed platform could support later. A lack of public evidence about current enablement should be labeled unknown, not treated as reassurance that the capability cannot be activated.

Why this can matter to protesters

Public video can document who attended a rally, who spoke with whom, what signs they carried, which vehicles arrived and where people went afterward. PTZ control and recorded footage make closer or later review possible. ALPR data can connect a vehicle to the same time and place.

This does not mean every camera near a demonstration is being used that way. It means the capability can chill lawful speech and association unless strict purpose limits are both written and enforced.

A U.S. Department of Homeland Security privacy assessment of public-camera systems identifies this general risk directly: cameras can document people at rallies and reveal associations, potentially chilling protected expression. It also warns about profiling based on characteristics including race, citizenship status, religion and socioeconomic status.

DHS privacy assessment: public video-surveillance risks →

Why this can matter to immigrants

The risk is not that a Condor camera can see immigration status. A camera cannot.

The risk is that identifiable video, vehicle observations, location, time and cross-agency requests can be used to locate a person, connect associates or document visits to homes, workplaces, religious institutions, legal services or community events.

Whether that happens depends on access, sharing, policy, law and practice. A vendor saying a federal agency lacks direct access does not answer whether another agency can search or share information on its behalf.

Local officials should identify every direct and indirect path by which immigration authorities or outside task forces could receive results, clips, exports, screenshots or live assistance.

Why this can matter to neighborhoods already watched more heavily

Camera and acoustic-sensor placement determines whose daily life is most often recorded or subjected to automated classification.

If equipment is concentrated in particular neighborhoods, residents there bear more of the false-alert risk, more routine police visibility and more collection of ordinary activity. Labels such as “high crime” do not remove the need to examine placement criteria, error rates, response patterns and disparate effects.

The Berkeley acquisition report itself acknowledges loss of anonymity in high-foot-traffic public spaces, possible disparate effects and the need to prevent discriminatory use. Those are not speculative objections invented by this campaign; they are standard surveillance-governance questions.

Why profiling does not require facial recognition

A system can help build a profile without knowing a person’s legal name from their face.

Repeated video, vehicle sightings, time, place, clothing, companions, travel direction and recurring visits may reveal patterns. Combining those observations can support inferences about where someone lives or works, whom they meet, which events they attend and which institutions they visit.

Some inferences will be wrong. That is part of the danger: a system can make an association easy to generate while leaving the affected person with little way to know about or correct it.

The questions Metro Detroit should answer before any purchase or renewal

Ask for records—not assurances—showing:

every Flock product, camera, sensor, software module and integration currently owned, leased, trialed or proposed;
device locations and the written rationale for each location;
contracts, order forms, amendments, renewal dates and termination deadlines;
product manuals and the exact features enabled locally;
retention settings for video, audio clips, alerts, metadata, exports and preserved evidence;
all user roles, access rules, sharing relationships and audit procedures;
written prohibitions or permissions involving protests, immigration enforcement, First Amendment activity and protected characteristics;
searches, alerts or referrals involving outside agencies, task forces or federal partners;
accuracy testing, false-alert records, dispatch outcomes and complaints;
cybersecurity incidents, exposed feeds, unauthorized access and vendor notices;
whether feature changes require council approval and public notice;
and whether deleting the system would also delete or return locally controlled data.

Do not assume an ALPR policy governs Raven or Condor. Each data type and integration needs to be named.

Our position

There can be legitimate reasons to investigate a shooting, review relevant video or use a camera for a specific security purpose.

The problem is building persistent, integrated surveillance infrastructure that watches ordinary public life first and waits for a reason to search it later.

Adding microphones, live video, person alerts and high-powered zoom to a plate-reader network does not make that concern smaller. It gives local residents more reason to demand the contracts, configurations, policies, audit records and end date.

Ask your local government to disclose every connected Flock product before the next purchase or renewal—and to decline or end systems that turn public life into a searchable surveillance platform.

Find the contract and renewal window →

Research your municipality →

Contact local officials →

Next: How Flock Nova connects records, personal information and camera networks →

Beyond Flock: How Leonardo SignalTrace follows wireless-device signatures →