Flock's Hot Lists are a Hot Mess

Flock’s platform allows agencies to use existing watchlists and to place license plates — and potentially other characteristics — on what it calls “custom hotlists.” Flock allows these hotlists to be freely exchanged between agencies without any meaningful limitations. Few, if any, laws seem to exist surrounding these watchlists, and few, if any, agencies have adopted policies about their use. There is no recourse for these largely secret, largely national watchlists.

Where Hotlists Come From

First, the official watchlist: this comes from the FBI’s National Crime Information Center (NCIC), which is “a computerized index of missing persons and criminal information and is designed for the rapid exchange of information between criminal justice agencies”.^[1] The FBI’s database was created in 1967, and has since grown tremendously in scope.

NCIC stores information about property (vehicles, firearms, etc.) as well as people (e.g., warrants, gangs, terrorism, anyone of interest to the secret service). This information can then be queried by state and local agencies, in accordance with the FBI’s terms, which are laid out in federal regulations and the CJIS Security Policy.

The information from NCIC — which is often inaccurate — is uploaded by states, after which the federal government and its state partners make it accessible to federal and local agencies via portals like Palantir’s ICM and Falcon, and, to the point of this article, Flock’s FlockOS and its “hotlists.”

Until recently,^[2] Flock had a blog post on its website called “What Happens When a Wanted Car Passes a Flock Safety Camera?” That post appears to have been deleted in a hurry. It is still linked to from Flock’s other blog articles, including one that describes the “Hot List feature.”

Because Flock has requested that sites like archive.org’s Wayback Machine not archive the content on its site, an archived copy is unavailable, but here is a paragraph from a referencing post:

Flock Safety’s Hot List feature allows businesses to receive alerts on stolen vehicles, known wanted criminals and more, as long as the private business chooses to share their LPR cameras with law enforcement. Law enforcement will then be automatically notified if a vehicle associated with the NCIC enters your property, allowing them to take action. It works via an integration into the Federal Bureau of Investigation’s National Crime Information Center (NCIC), which is updated daily.

Customers can also place at-risk vehicles or license plates on custom Hot Lists to alert law enforcement or on-site security instantly if/when they enter their property.

The “Customers” referred to are Flock’s commercial customers.

The Unregulated Layer

This is the second category of watchlist: managed by local police agencies, HOAs, neighborhood associations, and businesses, and shareable without constraints — but somehow still exempt from public records law, according to police agencies.

NCIC entries, which have specific retention policies and restrictions on sharing, are often copied to custom log entries in the Flock system.

Other entries, like the one shown on the “SUS Vehicles” hotlist, can be for vague reasons like “Poss vin swap, stop and verify” — suggesting the vehicle has been placed on the list to be stopped without probable cause, or, at best, for pretextual reasons.

These stops are then combined with departmental policies to keep their use secret, like those seen in Wapello County, Iowa:

DO NOT MENTION ALPR USAGE TO THE OCCUPANTS OF THE VEHICLE. Simply reference that you ran the plate and observed an NCIC hit … DO NOT MENTION ALPR USAGE IN YOUR REPORT OR COMPLAINT UNLESS ABSOLUTELY NECESSARY. … If it is necessary to explain in a report, it is advised to use language such as “Using County resources, I discovered the suspect vehicle was bearing Iowa plate…”

Whether a stop was conducted based on a watchlist hit can only be established if a defense attorney asks the question and the police answer it. That may not happen until the victim of an unlawful stop has spent days, weeks, or even months in jail.

What Counts as a “Reason”

I examined 800 “Custom Hotlist” entries entered by Harris County Constable Precinct 5^[3], which covers over 300 square miles of western Houston suburbs — an area with 1.3 million residents, policed by roughly 450 sworn deputies. 101 of those deputies — nearly one in four — appear as users in the hotlist logs. 51% of entries had no listed case number and 73% were set to never expire. The criteria used varied from outstanding warrants (copied over from NCIC), vague suspicion that a VIN may be altered, or no reason at all.

An investigator at the Precinct 5 constable’s office used Flock to locate and seize a vehicle for an alleged theft that another county refused to investigate, that was never entered into any criminal database, and whose statute of limitations expired seven years earlier. The hotlist entry reads:

Vehicle is not listed as stolen in NCIC/TCIC. Registered owner claims vehicle was stolen out of Colorado County in 2012. Colorado County refused to enter vehicle as stolen. Find PC and stop/identify all occupants… Vehicle will be seized for a Civil Seizure hearing. Do not release vehicle

Investigator James Dancer created this entry on July 3, 2024, for a Texas plate ending in –981^[4], case number 2407-00085, with a one-month expiry of August 2, 2024.

The vehicle was allegedly stolen twelve years ago, according to a claim by the registered owner. Another agency — Colorado County — had already declined to enter it as stolen; possibly because the statute of limitations expired seven years earlier, in 2017.^[5]

Colorado County already decided there was no criminal case here. Dancer’s entry instructs his peers to manufacture a justification for a stop (“Find PC”) and seize the vehicle anyway. That outcome (“Do not release vehicle”) is predetermined in the hotlist entry. Once the vehicle is seized, civil forfeiture requires only a “preponderance of the evidence” that the vehicle is connected to criminal activity.

The vehicle, which is at least twelve years old, won’t be returned to the registered owner. Instead, it will be sold at auction and the proceeds will go to Dancer’s department and the Harris County district attorney. The Institute for Justice has previously filed a class-action suit for exactly this type of abuse of the civil asset forfeiture process in Houston.

Dancer’s entry also contains an instruction that appears 47 times across the 800 entries: “Find PC and stop/identify all occupants.” In 31 of those, the instruction was created by Kayla Cohan (formerly Fesperman) using a near-identical template: “BMV Susp Vehicle- BLK FORD BRONCO-Develop PC, Stop and ID Occupants.” Burglary of a Motor Vehicle (BMV) is typically a misdemeanor. These entries instruct officers to first locate a vehicle through Flock’s automated surveillance, then find or develop a legal pretext for the stop afterward — inverting the Fourth Amendment’s requirement that probable cause precede the seizure.

There is no reason to believe that the “occupants” of these vehicles, who are entirely unknown at the time the entry is created, are involved in the crime the driver is suspected of committing. That crime is equally unknown until someone “finds PC.”

Similarly, another one of Dancer’s entries for a vehicle that could be related to shoplifting at Bath and Body Works contains a slightly more detailed instruction:

Bath and Body Works Shoplifting Suspect Vehicle/ Find PC, Stop and Identify all Occupants. if any arrests are made, contact Investigator James Dancer (5I10), will most likely make scene. Contact Pct. 5 Dispatch (281) 463-6666. Send all questions/ information to James.Dancer@cn5.hctx.net

The investigator email addresses and direct phone numbers embedded in these reason fields serve a secondary purpose: when hotlists are shared between agencies, the receiving agency also receives internal contact information for the creating agency’s investigators — an informal contact-sharing network piggybacking on the surveillance system.

Another vehicle is put under indefinite surveillance by Andrea Trevino. No case number given: “SURVEILLANCE ONLY//DO NOT MAKE CONTACT.”

Jose Ramos added 6 vehicles for “Tolls” — five with no case numbers, one with case number “N/A”, all set to never expire.

Mental health crises can land you on the watchlist: “SUICIDAL GUY” (one month), “WARRANT AND SUICIDAL GUY” (no expiration), “SUICIDAL ARMED” (one month), “Welfare Check” (one month). None had case numbers.

You can also earn a permanent watchlist spot for how your car sounds: “Engine does not sound stock;” or for cryptic reasons like “300 has a badge”, “Memorial Mall”, or simply because the sergeant said so (“Sgt request”, case number: “theft warrant”).

Other entries include “SUS”, “sus veh”, “susp”, and “fug”.

A dashboard from an Iowa agency shows that even where reasons are entered, they indicate problems; there is a hotlist for “Sex Offender” — a conviction is not justification for indefinite surveillance — as well as a hotlist for “Protection Order.”

That latter category’s existence is surprising, given that Arkansas and Wisconsin police failed to act when a person with an ankle-monitor passed at least one Flock camera, as he went to the home of a 16-year-old who had a protective order in place. The pair then disappeared for two months, until someone spotted them at a truck stop in Nebraska.

Sharing Without Scrutiny

When hotlists are shared, the receiving agency receives an automated email from Flock. That email considers the data sharing to be “Great News!” and provides instructions on how to start using the shared hotlist.

What it does not provide is information about what is on the list — including whether the hotlist contains any information sourced from NCIC, which would be restricted by law. It also does not provide information about the policies in place for maintaining and populating the list.

Flock sent the email above to a Minnesota agency. The email does not mention that Minnesota law^[6] says ALPR data “may only be matched with data in the Minnesota license plate data file” or when related to an active criminal investigation. Instead, Flock cheerfully directs its agency customer to accept the mystery hotlist from Florida — in a way that more likely than not violates Minnesota law — and thanks them “for being part of the Flock.”

Although Transparency Portals include a “Hotlists Alerted On” column, Transparency Portals don’t appear to show custom hotlists.^[7] Flock’s transparency portal configuration tool, however, shows that “This value will be automatically generated according to your Flock settings.”

Flock Won’t Say How Often It’s Wrong

A common concern surrounding the use of automated enforcement is accuracy — we know that misreads happen, but we don’t know how often. And Flock won’t say.

Flock monitors and collects misread rates; it declined to provide Business Insider with specifics about the data. When customers flag misreads, that data is pulled into the company’s training set to improve its model, and the company works with local law enforcement to understand the cause of the incident, a spokesperson said.

What Flock has said, in its undated^[8] blog post “Assessing the Accuracy of Computer Vision Methods for Traffic Data Collection,” is that “Flock Safety cameras correctly identified 92.3% of vehicles by classifying them across six vehicle categories via computer vision, in accordance with the FHWA’s 13-bin classification system.”

The FHWA 13-bin system uses broad categories like “Motorcycle” or “Five-axle single-trailer trucks.” It has several categories for tractor-trailer combinations with differing numbers of axles. Flock doesn’t specify, but most likely reduced the number of bins to six by collapsing several vehicle classes into a single “truck” classification.

That 92.3% figure measures vehicle type classification — whether the system can tell a motorcycle from a tractor-trailer — not plate-reading accuracy. If Flock’s cameras fail to correctly classify vehicle type in nearly 8% of cases using these broad categories, their accuracy at reading individual plate characters or matching the more granular attributes in its FreeForm search — clothing, dents, and other vehicle and person characteristics — is anyone’s guess. Flock’s refusal to disclose plate-read accuracy or submit to independent auditing leaves that question unanswered.

Even NCIC Gets It Wrong 77% of the Time

But even when accuracy is discounted, an overview of Axon data from the Story County, Iowa sheriff’s office^[9], shows that matches are often a bust. In Story County’s “Erroneous hotlist hits” report, which covers approximately a month, hits were only sourced from NCIC.

In that month, the sheriff’s office reported 214 incorrect hits. 165 (77%) were flagged “Wrong state,” 10 were “correct” but “No action taken,” “No Associated Party in Vehicle” or “Parked - Unoccupied,” 7 were “Dismissed,” and 3 were “incorrect.” The report does not include information about the total number of scans, or the number of accurate hits.

77% of vehicles from NCIC being matched to vehicles from the wrong state is an issue. The issue is further compounded by NCIC not distinguishing between characters like “O” and “0” or “I” and “1”. And that’s NCIC, which has policies on accuracy and review of information.^[10] For Flock’s “Custom Hot Lists,” all it takes for someone in Minnesota to get pulled over is for Brittany Smith from Florida to enter an incorrect digit.

Even if the technology were 100% accurate, which it isn’t, it is still subject to the principle of “Garbage in, garbage out.”

Who Else Gets the Data

After an entry is created or copied into a custom hotlist, users can opt to receive notifications through Flock’s mobile app. If the entry is criminal justice information (CJI) copied from NCIC, it may only be accessed through a secure workstation.

Instead, Flock encourages use of its mobile app, which offers simple on/off toggles for receiving alerts. It also offers notifications based on a user’s location, suggesting Flock mobile app users’ locations are being recorded and tracked.

The risks of inaccurate or poorly-maintained hotlists to the public are well-documented. In Sherwood, Arkansas, a Flock ALPR misread a plate obscured by a loose plate holder. Police ordered a couple out of their car at gunpoint and handcuffed the woman in front of her children. In Morristown, Tennessee, a Flock camera misread an “O” as a “0,” and the Herron family — with their 3-year-old granddaughter in the car — was pulled over at gunpoint. In Toledo, Ohio, a misread “7” became a “2,” and Brandon Upchurch was mauled by a police dog and jailed. And in Aurora Colorado, an unidentified system flagged an SUV as a stolen motorcycle from another state — the plate number was the same, but it was the wrong state, like in 77% of Story County, Iowa’s erroneous hits.

But members of the general public are not the only ones at risk — police officers are too.

The hotlist data itself is federally regulated under the CJIS Security Policy:

[W]hether it’s bring your own device (BYOD) or a agency-issued phone, [law enforcement] must follow the protocols set out by the FBI’s Criminal Justice Information Services (CJIS) compliance for mobile device security and adhere to the CJIS MFA requirements that go into effect October 1, 2024

But those rules don’t extend to the data Flock collects through the app — like location data. It can collect or resell that data, or use it in its other products. There is no requirement for Flock to safeguard that data, or to screen the people — employees and contractors — with access to that data.

How much of the data Flock transmits is unclear, but the mobile app has a direct integration with FullStory^[11], a “Behavioral Data & Digital Analytics Platform.” It is equally unclear where the data goes from there.

We have already seen the predictable outcome when Flock failed to secure an API key and leaked live location data, including this exact “Officer mobile app location data (phone, smartwatch).”

With no functional safeguards in place, a history of leaks, and no consequences imposed by the Department of Justice or state agencies, it is left for individual officers to decide if they want Flock to enter their realtime locations — potentially both on-duty and off-duty — into its database, and whether all of Flock’s employees, contractors, and partners should have access to that data.

No Recourse, No Oversight, No End Date

There is no mechanism to discover which agencies are using what custom hotlists, or for a person to discover whether they are on a custom hotlist, to challenge their inclusion, or to request removal. There is no judicial review, no expiration requirement, and no independent audit.

The entries examined here — 73% set to never expire, half with no case number, many with reasons no more specific than “SUS” — are not aberrations in an otherwise functional system. They are the system functioning as designed.

Flock’s marketing promises transparency and accountability. Its logs consistently tell a different story: indefinite surveillance authorized by a single officer’s keystroke, shared across jurisdictions without review, immune from public records requests, and enforced through pretextual stops that its own users are instructed to manufacture.

The question is not whether the technology works. The question is whether anyone is watching the people who use it.

The answer, to date, is “No.”