Flock Surveillance Cameras Now Track People, Not Just Cars

What Flock Safety Actually Is — and What It Has Become

Flock Safety is an Atlanta-based surveillance technology company founded in 2017. Its core product is a fixed-mounted Automatic License Plate Reader (ALPR) camera that reads passing license plates and flags stolen vehicles or cars tied to active warrants in real time. That framing has made Flock politically easier to deploy than more contentious technologies like facial recognition: the company has grown into one of the most widely used surveillance vendors in American law enforcement, with cameras installed by thousands of agencies and homeowners' associations across the United States.

But Flock's product portfolio has expanded well beyond plate readers. Its Flock Video Cameras line goes considerably further than ALPR: per Flock's own product documentation, these cameras include "People Detection Alerts" and a "Guardian Mode" that surface both vehicles and pedestrians moving through the frame, alongside PTZ hardware with high-magnification zoom. Layered on top of this hardware is FreeForm, an AI-powered search interface Flock lists as part of its platform. Flock markets FreeForm heavily around vehicle and footage searches, but the feature's actual architecture — natural-language queries interpreted by Flock's AI and computer-vision stack — makes it equally capable of surfacing people by physical description.

According to reporting by Joseph Cox for 404 Media, police departments across the United States are using this search feature to look for people by their clothing, tattoos, hair color, body type — and, in some cases, references to their apparent race — not merely the license plates the company publicly advertises. The outlet reports, based on records it reviewed, that officers have used Flock's FreeForm search tool at least hundreds of times, with individual searches able to reach many hundreds of cameras at once. The implications for civil liberties, algorithmic accountability, and the regulation of AI-powered surveillance infrastructure are significant — and largely invisible to the public.

What this means in practice: an officer can type a sentence into a search bar and, according to 404 Media's reporting, scan footage drawn from large numbers of participating camera networks — all without the public debate that would accompany the rollout of, say, a citywide facial-recognition system.

How FreeForm Works: A Search Engine for Surveillance Footage

FreeForm lets officers query Flock's federated camera network the way someone might use a search engine — in plain, natural language. The platform's AI interprets the phrase, scans indexed footage across participating networks, and surfaces clips matching the description. 404 Media reports that a single FreeForm query can reach hundreds of individual cameras at once, and that some searches spanned numerous separate camera networks simultaneously — networks that may each contain multiple physical camera units.

The scope is hard to overstate. This is not an officer pulling local CCTV footage and manually scanning it. It is a federated, AI-mediated search that can cross city, county, and state lines quickly — operated through a text box by any credentialed officer with platform access.

Documented Real-World Searches

The 404 Media report describes a striking variety of people-focused searches. The examples below reflect query strings the outlet reported from the records it reviewed. Where a query's exact reach was not published, that is noted; where the source described reach in individual cameras versus camera networks (each network may include multiple cameras), the distinction matters and should not be conflated.

Type of query reported	Example (as reported by 404 Media)	What it targets
Clothing description	"person wearing orange vest and construction hat"	A person by attire
Activity / posture	"person on skateboard"	A person by behavior
Physical description + clothing	"heavy-set male with a black and white hat"	A person by build and attire
Race/appearance references	Queries that referenced a subject's apparent race	A person, in part by demographic attribute
Political-affiliation references	Queries that referenced visible signs of political affiliation	A person/vehicle, in part by political symbol

According to 404 Media, several searches went beyond clothing or body type to reference a subject's apparent race, and at least one referenced visible political affiliation. Because the underlying records and precise per-agency figures were reported behind the outlet's paywall and are not independently verifiable here, this article treats those specifics as 404 Media's reporting rather than as independently confirmed facts. What is clearly reported is the pattern: the tool is being used to find people, not just vehicles, and race- and affiliation-adjacent descriptors have appeared in real queries — raising immediate questions about the guardrails Flock says it has built into the system.

Flock's "Guardrails" — and Why the Reporting Raises Doubts

Flock Safety has publicly positioned FreeForm as a responsible, limited tool. The company's stated position is that FreeForm is designed to help investigators search large amounts of footage when working with limited information — such as a witness description of a person or vehicle — and that it is not facial recognition. Flock states that it has no facial recognition technology in its products or in development, and that FreeForm cannot identify a person by name, verify someone's identity, or search for a specific face.

Flock has also described technical guardrails intended to prevent searches by characteristics such as race, ethnicity, religion, and nationality, and has said that attempts to run such searches can trigger alerts to the querying agency's administrators.

404 Media's reporting sits in tension with that framing: the outlet reports that queries referencing a subject's apparent race nonetheless appear in the records it reviewed. Several explanations are plausible — the guardrails may not reliably detect race references embedded in multi-attribute, natural-language queries; they may flag such queries while still returning results; or they may be scoped narrowly enough that race used as one attribute among several is not caught. None of those scenarios is reassuring, and, as of this reporting, it is not publicly clear which applies or whether the described searches triggered any administrator alerts.

The distinction Flock draws — FreeForm is not facial recognition — is technically meaningful but practically narrow. The system does not match a biometric faceprint against an identity database. But it does enable mass, AI-assisted scanning of large numbers of cameras to locate a specific individual matching a physical or demographic description, producing a functionally comparable surveillance outcome while carrying far less legal and political weight than the term "facial recognition" attracts.

The Political Affiliation Problem and Broader Scope Creep

Among the most constitutionally charged patterns 404 Media reports is the use of FreeForm to find people based, in part, on visible political affiliation. Unlike a license plate search — anchored to a specific registered vehicle — a query built around a political symbol effectively targets a person in part because of protected expression. The First Amendment implications of law enforcement using an AI-powered, cross-jurisdictional search to locate someone based on political insignia have not been tested in court, but they represent a genuinely novel legal frontier.

Other reported searches extend the scope into territory that would have seemed far-fetched five years ago. Queries such as "person on skateboard" and "person wearing orange vest and construction hat" — drawn from 404 Media's reporting — suggest the bar for initiating a broad camera search can be remarkably low and informal, and that officers are using FreeForm not merely as a reactive forensic tool after a crime is reported, but as a proactive visual-identification system that scans public space for behavioral or visual attributes.

Legitimate Use Cases, Missing Warrants, and the Oversight Vacuum

Flock has cited genuine public-safety scenarios in defending FreeForm, and such use cases are real and compelling. Locating missing children in the context of an AMBER Alert, or finding a vulnerable adult who has wandered from care, are exactly the kinds of examples that make it politically difficult for legislators to impose blanket restrictions on the technology, and that give agencies a ready answer when pressed by reporters or oversight bodies.

404 Media reports that some searches are formally logged as part of an active investigation, while others are categorized as missing persons cases — but that a significant number of the stated reasons in the public records are redacted, making independent oversight of the full scope of use difficult.

This redaction gap sits at the center of the oversight problem. Consider the legal frameworks governing other tools police use to locate and track people. Do cops use GPS trackers to follow suspects? Yes — and in United States v. Jones (2012) the Supreme Court held that physically attaching a GPS device to a vehicle and using it to monitor movements is a Fourth Amendment "search"; the majority rested on the physical trespass, while concurring justices emphasized the privacy intrusion of long-term tracking. Do cops use grappling hook or grappler devices to stop fleeing vehicles? Those tools — like the Grappler Police Bumper — typically come with use-of-force policies, supervisor approval requirements, and documented deployment protocols. Can cops use Flock cameras through FreeForm with the same level of procedural rigor? Based on publicly available evidence, the answer appears to be no: FreeForm queries do not appear to carry a universal warrant requirement, and in many reported cases there is little indication of meaningful supervisory review before execution.

Why it matters: The legal question is not whether FreeForm is facial recognition. The question is whether a technology that lets a single officer quietly scan large numbers of cameras across multiple jurisdictions for a person matching a physical or demographic description — without a warrant, without judicial oversight, and without meaningful public disclosure — is consistent with a reasonable expectation of privacy in public space.

Courts have been slow to catch up. The third-party doctrine — which holds that information voluntarily exposed to third parties may carry no Fourth Amendment protection — may shield many FreeForm searches from challenge under existing precedent. But the Supreme Court's 2018 ruling in Carpenter v. United States, which required a warrant for historical cell-site location data despite its third-party origins, signaled that the sheer volume and comprehensiveness of surveillance can itself trigger constitutional scrutiny, even when each individual data point seems innocuous. The Court framed that holding narrowly, but a system that aggregates footage from many cameras across multiple jurisdictions in response to a single query looks far closer to the "detailed, encyclopedic, and effortlessly compiled" surveillance Carpenter worried about than a single officer checking one local camera.

Frequently Asked Questions About Cops and Flock

Can cops use Flock cameras to track individuals without a warrant?

Based on current publicly available information, there is no documented universal warrant requirement for FreeForm queries. Individual agencies may have internal policies, but no federal statute or widely adopted national standard currently mandates judicial authorization before an officer runs a person-description search across Flock's network. This is an area of active legal uncertainty.

Do cops use Flock cameras for facial recognition?

Flock states that FreeForm is not facial recognition and that its systems cannot identify a person by name or match a specific face to an identity database. That said, the tool's ability to scan large numbers of cameras for someone matching a detailed physical description — including references to build, hair color, clothing, and in some reported cases apparent race — produces a surveillance outcome that civil-liberties advocates argue is functionally analogous, even without biometric face-matching.

Do cops use GPS devices alongside tools like Flock?

Yes. GPS tracking devices remain a separate, widely used law-enforcement tool subject to the Fourth Amendment "search" framework established in United States v. Jones (2012). FreeForm and GPS can serve overlapping investigative purposes — locating a specific person — but operate under very different legal frameworks, with GPS attachment facing clearer judicial constraints than FreeForm currently does.

What is the difference between Flock's ALPR cameras and its video cameras?

Flock's standard ALPR cameras are fixed, plate-reading devices designed primarily to flag vehicles. Flock's separate video camera line adds capabilities such as People Detection Alerts and Guardian Mode, capturing pedestrians as well as vehicles, and includes PTZ models with high-magnification zoom. FreeForm can search footage from Flock's platform, and the video cameras' people-detection capabilities make them a stronger foundation for person-focused searches than plate readers alone.

What This Means for Developers and the AI Surveillance Stack

For the technically literate reader, the FreeForm story is as much about system architecture as civil liberties. Flock has built what is, in engineering terms, an impressive piece of infrastructure: a federated camera network with a unified natural-language query layer, backed by computer-vision models capable of parsing free-text descriptions into visual search parameters at scale across many endpoints. That is a genuinely difficult distributed-systems and ML problem, and the product appears to work effectively.

The trouble is that the same architectural properties that make FreeForm powerful — federation across many independently operated networks, natural-language abstraction over visual search, real-time cross-jurisdictional reach — are precisely what make it dangerous without robust governance. When you abstract a broad search behind a consumer-friendly text box, you reduce the cognitive and procedural friction of conducting mass surveillance to nearly zero. An officer who might hesitate before formally requesting weeks of archived footage from dozens of separate jurisdictions — a process that would require paperwork, supervisor approval, and inter-agency coordination — faces little of that friction when typing a sentence into FreeForm.

This friction-reduction pattern is worth watching across the broader AI surveillance stack. Tools like FreeForm will likely become more accurate, more granular in the attributes they can parse, and more widely deployed as underlying vision models improve. The gap between what these systems can do and what legal frameworks permit — or prohibit — will keep widening unless legislators and courts act with unusual speed and technical specificity.

Developer ethics also enter the picture in a way the industry has largely avoided confronting. When a company builds a capable AI system and sells it to law enforcement, the downstream consequences of misuse or scope creep are not solely a customer-configuration problem. Flock designed FreeForm's query interface, trained its vision models, and built the federation layer connecting many independent networks. If the system processes race-referencing queries — whether or not the stated guardrails reliably intercept them, as 404 Media's reporting suggests they may not — that reflects a product-design reality embedded in the architecture, not merely an end-user error. The company's responsibility for those design choices doesn't disappear behind a terms-of-service clause.

Key Takeaways

• Cops use Flock's FreeForm feature to track people — not just vehicles — using natural-language queries covering clothing, activity, body type, and, per 404 Media's reporting, references to apparent race.
• 404 Media reports FreeForm has been used at least hundreds of times across law-enforcement agencies, and that individual queries can reach hundreds of cameras at once, including networks operated across jurisdictions.
• Reported searches include descriptions referencing a subject's apparent race and, in at least one case, visible political affiliation — patterns that appear to run against Flock's stated guardrails restricting demographic queries.
• Precise per-agency query strings and reach figures were reported behind 404 Media's paywall and are treated here as the outlet's reporting rather than independently verified records.
• Flock states FreeForm is not facial recognition — but the functional outcome of locating a specific person across a large, cross-jurisdictional camera network is comparable in practical effect, while facing fewer legal and political restrictions.
• There is no documented universal warrant requirement for FreeForm queries, and 404 Media reports that a substantial portion of search justifications in public records are redacted, hampering independent oversight.
• Flock's video camera line (distinct from its ALPR hardware) includes People Detection Alerts and Guardian Mode — features that go beyond the plate-reading product Flock primarily markets to the public and policymakers.
• Legitimate use cases — AMBER Alerts, missing vulnerable adults — are real, but they don't resolve the oversight vacuum surrounding the broader range of searches now surfacing in the public record.

What Comes Next

The immediate pressure point is legislative. Several states have enacted laws regulating ALPR data retention, access, and sharing, but such statutes were generally written to govern what happens to plate data after collection — not a natural-language, AI-mediated person-search system like FreeForm. They say little about a query interface that searches footage for people by clothing, tattoo, or apparent race. Advocacy organizations are likely to use 404 Media's records as an evidentiary foundation for legislative testimony and potential legal challenges. Civil-liberties groups such as the ACLU have been building strategies around algorithmic surveillance for years, and a documented record of race-referencing searches that appear to have bypassed company-stated guardrails is exactly the kind of concrete evidence that advances those efforts in court and in legislative hearings.

For Flock Safety, the reporting creates reputational and regulatory exposure that its ALPR-forward public framing can no longer contain. The company may face calls to testify before state legislatures and potentially Congress, and any gap between its guardrail claims and the reported search data will be a centerpiece of that scrutiny. Whether Flock responds by hardening its query filters, restricting FreeForm access to a narrower set of approved use cases, building in warrant-requirement integrations, or defending its existing law-enforcement partnership model will be a significant indicator of where the commercial AI surveillance industry believes its accountability obligations begin and end.

For the public, the story's most durable lesson may be the simplest: the surveillance infrastructure most likely to shape daily life is rarely the dramatic system announced with press releases and public debate. It is the quietly expanded feature that ships inside the product you already accepted. Flock's cameras were introduced as license-plate readers. The cameras stayed. The capabilities grew. And by the time the public record caught up, 404 Media reports, hundreds of people-searches had already been run across large numbers of camera networks — with no warrant, minimal oversight, and little public awareness.