Flock Exposed - Misidentification and Wrongful Detention
License Plate Reader Errors Lead to Gunpoint Stops
The Institute for Justice, a nonprofit legal foundation, has documented at least 26 to 27 cases since 2018 of innocent motorists pulled over, detained at gunpoint, or jailed because of automated license plate reader errors, with most of these cases occurring since 2022 or 2023. The organization notes this count is most likely an undercount, since there is no comprehensive national database of ALPR errors.
Source: Institute for Justice, "Dozens of Innocent Motorists Have Been Pulled Over, Detained at Gunpoint, or Jailed Due to AI License Plate Camera Errors"
In February 2026, a Flock camera in Sherwood, Arkansas, misread the license plate of an SUV, leading officers to detain an innocent couple at gunpoint while their six-week-old baby sat alone in a car seat in the back of the vehicle.
Source: Institute for Justice
In Brookfield, Wisconsin, police pulled over driver Amber Newell after a Flock camera flagged her car in connection with a Milwaukee homicide. She was ordered out of her car at gunpoint with her hands up. She was stopped at gunpoint twice within one week; the alert should have been cleared days earlier, but staff failed to remove it.
Source: Institute for Justice, via IJ Update; Shasta Unfiltered
A Florida woman, 23-year-old Lindsey Isaacs, spent seven months dealing with a wrongful arrest after police used a Flock camera to identify what turned out to be the wrong car entirely.
Source: Reason magazine
In one case documented by researchers, a driver was pulled over at gunpoint on his way to work and handcuffed face-down on suspicion of driving a stolen vehicle, after an ALPR system misread the letter H as the letter M on his license plate. The case settled for thirty-five thousand dollars in April 2025; during the settlement conference, the federal judge said that "Flock Flocked up," according to the driver's attorney, Peter Pattakos.
Source: ObscureIQ, "ALPR Accuracy: The Misread Problem"; AOL, "AI cameras are everywhere - and people are paying the price for their mistakes"
In San Diego in November 2025, officers searched Flock's system by vehicle make and model rather than by license plate while looking for a red Alfa Romeo. The system matched an entirely different, identical-looking car five miles away. Two men were arrested at a cigar lounge, and one was jailed for nearly a month before charges were dropped. They are now suing the city for false arrest, wrongful detention, and negligence.
Source: apex-cameras.tech ALPR camera facts summary
In San Diego, the man jailed for nearly a month in the cigar lounge case, Hugo Parra, could not afford bail while his co-defendant, Beltran, was able to post bail and was released. Lawsuit documents note the exculpatory data - showing the real car was five miles from the crime scene - existed inside Flock's own ALPR system at the time of Parra's arrest. Charges were ultimately dropped.
Source: 10News San Diego; Times of San Diego, via Slashdot
In Volusia County, Florida, in April 2026, a driver was jailed for thirteen days after a Flock camera capture was mistakenly matched to a vehicle involved in a fatal accident.
Source: apex-cameras.tech ALPR camera facts summary
Privacy advocate Brian Hofer and his younger brother were held at gunpoint by law enforcement in 2018 while driving to visit their parents for Thanksgiving, after Hofer's vehicle was incorrectly flagged as stolen by an ALPR system. Hofer, who now leads the organization Secure Justice, settled his case for 49,500 dollars and has said his life was different afterward; he advocates for manual verification of ALPR hits before any armed response.
Source: Consumer Rights Wiki, citing CBS News reporting; Electronic Frontier Foundation, "The Human Toll of ALPR Errors"
In Espanola, New Mexico, a 12-year-old girl named Jaclynn Gonzales was handcuffed after a license plate reader camera misread the last digit of a plate on a vehicle driven by her older sister, mistaking a "2" for a "7," according to a lawsuit filed against the city.
Source: CBS News, "When license plate readers get it wrong"; Consumer Rights Wiki
One month later in the same city, 17-year-old honor-roll student and starting high school quarterback Adam Pacheco was held at gunpoint on his way home from school after an officer misidentified his pickup truck as one connected to armed robberies in a be-on-the-lookout alert. He is now suing over the stop.
Source: Santa Fe New Mexican, "Lawsuit: Police mistakenly held Pojoaque teen at gunpoint"; KRQE News 13
In a 2009 case that predates Flock itself but established the same pattern later seen nationwide, San Francisco police pulled Denise Green out of her car at gunpoint, handcuffed her, and searched her vehicle after an ALPR system misread a "3" as a "7" and flagged her car as stolen. Officers had not visually verified the plate before the stop. Green's civil rights lawsuit was initially dismissed, then reinstated by the Ninth Circuit Court of Appeals, and the city ultimately paid her 495,000 dollars in settlement.
Source: Electronic Frontier Foundation, "The Human Toll of ALPR Errors"; Courthouse News Service
An LAPD Inspector General audit found that roughly one in three Flock hot-list alerts reviewed were inaccurate. Los Angeles subsequently let its Flock contract expire in July 2026.
Source: AI Weekly, citing the LAPD Inspector General audit
Beyond wrongful stops, the Institute for Justice has cataloged more than two hundred incidents of ALPR abuse altogether, including at least 47 documented cases of officers using license plate reader databases to stalk romantic interests, such as a Louisville, Kentucky officer who tracked the mother of his child for months, running more than two hundred searches on her after she obtained a protective order against him.
Source: Institute for Justice ALPR Abuse Database; IJ Update
Facial Recognition Misidentifies People of Color at Far Higher Rates
A landmark 2019 study by the National Institute of Standards and Technology tested 189 facial recognition algorithms and found that Black and Asian faces were misidentified at rates ten to one hundred times higher than white faces.
Source: National Institute of Standards and Technology, cited by State of Surveillance
A UK Home Office study similarly found that Black and Asian people are more likely to be incorrectly matched by facial reconition systems than white people.
Source: State of Surveillance
The Innocence Project counts at least seven confirmed cases in the United States of people wrongfully accused because of facial recognition misidentification, six of whom are Black: Robert Williams, Michael Oliver, Nijeer Parks, Randal Quran Reid, Porcha Woodruff, and Alonzo Sawyer.
Source: Innocence Project
The ACLU has separately documented more than a dozen wrongful arrests tied to facial recognition, including Robert Williams (Detroit, January 2020), Christopher Gatlin (St. Louis, August 2021), Alonzo Sawyer (Maryland transit police, March 2022), Randal Quran Reid (arrested in Georgia on a warrant from Louisiana, November 2022), Porcha Woodruff (Detroit, February 2023), Jason Killinger (Reno, September 2023), Robert Dillion (Jacksonville Beach, Florida, August 2024), Javier Lorenzano-Nunez (Phoenix, October 2024), Trevis Williams (New York City, April 2025), Angela Lipps (arrested by U.S. Marshals in Tennessee on a warrant from Fargo, North Dakota, July 2025), and Beau Burgess (Orlando, August 2025).
Source: American Civil Liberties Union, "More than a Dozen Wrongful Arrests Due to Police Reliance on Facial Recognition Technology"
Robert Williams was wrongfully arrested outside his own home in Detroit in January 2020, in front of his two young daughters, his wife, and his neighbors, and was held for thirty hours in an overcrowded detention facility, based on a flawed facial recognition match. His case led to a settlement with the city.
Source: American Civil Liberties Union, "Williams v. City of Detroit"
Porcha Woodruff was eight months pregnant when Detroit police arrested her for carjacking and robbery based on a facial recognition hit, despite nothing in the surveillance footage or witness statements indicating a pregnant woman had been involved. She was detained for eleven hours and suffered contractions from the stress of the arrest.
Source: Humanoid Liability legal guide; Michigan Public radio
Randal Quran Reid was arrested by Louisiana-connected police based solely on a Clearview AI facial recognition match. Officers did not verify his location or check whether he had ever been to the state where the crime occurred.
Source: State of Surveillance, "Arrested by Algorithm: 13 Cases Dismissed, Nearly Every Victim Black"
An East Tennessee grandmother, Angela Lipps, was arrested by U.S. Marshals in Tennessee in July 2025 and held in custody for nearly six months, first in a Tennessee jail and then in North Dakota, after AI facial recognition software misidentified her as a bank fraud suspect wanted by Fargo police. She lost her home, her car, and her dog during her detention and was released on Christmas Eve with no way to get home. She is now suing the City of Fargo and the detective involved for 10 million dollars.
Source: TechSpot; WATE; ABC News
Several of these wrongfully arrested individuals had obvious physical differences from the actual suspect that a human reviewer might have caught: Michael Oliver had full tattoo sleeves the suspect did not, Porcha Woodruff was visibly eight months pregnant, and Trevis Williams was eight inches taller and seventy pounds heavier than the person police were looking for.
Source: ACLU of Oregon
Motorola's Vigilant Bundles Facial Recognition Directly Into Its License Plate Readers
Motorola Solutions is a major Flock competitor in the license plate reader market through its Vigilant product line. Unlike Flock's public claim that its own cameras do not use facial recognition, Motorola's Vigilant ALPR systems come integrated with facial recognition technology and an image database as part of the same product sold to law enforcement agencies.
Source: American Friends Service Committee, AFSC Investigate database, Motorola Solutions company profile
Flock's Own Patent Describes Face Recognition Searching
Flock's public trust pages state flatly that its cameras do not use facial recognition technology. In the company's own words: "Flock cameras do not use facial recognition technology. Searches are based on vehicle characteristics, and not identity. Flock cannot recognize, identify, or track individuals."
Source: flocksafety.com Trust page; flocksafety.com Civil Liberties and Rights Safeguards page
Flock's own patent filings describe something different. United States Patent 11,416,545 B1, titled "System and method for object based query of video content captured by a dynamic surveillance network," is assigned to Flock Group Inc. It is a continuation of an earlier patent, United States Patent 10,795,933 B1, with the underlying invention assigned to Flock Group Inc. in 2018 by its two named inventors, Matt Feury and Garrett Langley. Garrett Langley is Flock Safety's founder.
Source: United States Patent and Trademark Office records, as published on Google Patents, patents US11416545B1 and US10795933B1
The patent document itself describes searching surveillance footage for a particular human being by face. In the patent's own language, using "face recognition data points," investigators "may be able to query the DSOQ system for any footage captured by HSS-1 or HSS-2 having a person with matching face recognition data points."
Source: United States Patent 11,416,545 B1, specification text, as published on Google Patents
The walkthrough example at the center of that patent shows the intended scope. A car is spotted near a crime scene. Investigators trace that same car through an unrelated traffic camera, then through a convenience store's private security footage, and then use a face captured in that footage to search a third and separate network of home security cameras. Three independently owned camera systems, never built to talk to one another, stitched into a single timeline by the software and fed into a neural network.
Source: United States Patent 11,416,545 B1, as published on Google Patents; Clay Views, "Flock Says It Doesn't Track You, But Its Patents and Products Say Otherwise"
A second patent held by the same company, United States Patent 11,030,892 B1, "Method and system for capturing and storing significant surveillance images," was filed by Flock Group Inc. on February 10, 2020 and granted on June 8, 2021.
Source: United States Patent and Trademark Office records, as published on Google Patents, patent US11030892B1
In fairness to the record, when journalists at WIRED and 404 Media examined the software recovered from a physically opened Flock camera in September 2026, they found no evidence of active facial recognition capabilities in that software beyond features included by default in the Android operating system the camera runs on. Nothing in the public record proves that facial recognition is switched on in any camera on a pole today.
Source: IBTimes UK, "Flock Camera Captured 1.6M Images in 21 Days as Leaked Code Shows It Detects People," reporting on the joint WIRED and 404 Media analysis
That distinction is the whole point, and it cuts in a direction the company's marketing does not address. Flock's denials describe what the system currently does, not what it is capable of doing. A capability designed into the architecture, and claimed in a granted patent, does not require any change to the hardware on the pole in order to be activated. The equipment a town votes to install is not the same thing as the software that will be running on it a year later, and nothing in Flock's public statements commits the company to never turning that capability on.
Source: Analysis of the patent record above, read against Flock Safety's published trust and privacy statements; DeFlock BCS, "Flock Safety Says Its Cameras Don't Do Facial Recognition, So Why Does Its Patent Describe It"
The company has continued to make the narrower claim publicly. In a May 2026 blog post, Flock wrote that its cameras "do not use facial recognition" and "are not general-purpose surveillance systems designed to identify people." In June 2026, Flock's Chief Communications Officer, Josh Thomas, denied that the Flock system tracks people during an interview with Atlanta News First Investigates.
Source: The Daily Caller, September 17, 2026, "Surveillance Patent Shows Flock Spying Way Worse Than Advertised"
Flock's Condor pan-tilt-zoom camera line has already been shown capable of tracking a specific person in real time. During a June 2026 demonstration documented by the television investigative program InvestigateTV, a Condor camera mounted on a public trail pivoted, followed, and zoomed in on a passing reporter, even as a Flock Safety representative standing nearby was being told by that same reporter that the cameras do not track people.
Source: InvestigateTV, via Yahoo News and Gadget Review, "Flock Said Its Cameras Don't Track People. Then a Reporter Proved Them Wrong On Video"
Separately, leaked code examined by IBTimes UK showed that a Flock camera's own onboard models detected and identified a reporter in a test image, which was then run across more than 27,000 short video clips stored on the device, demonstrating that person-detection capability already exists at the camera level regardless of what any given customer contract enables.
Source: IBTimes UK, "Flock Camera Captured 1.6M Images in 21 Days as Leaked Code Shows It Detects People"
Questions to Ask Your Town
What is our own police department's documented error or false-positive rate for Flock alerts, and has anyone ever audited it independently the way the LAPD Inspector General did?
What policy requires an officer to visually verify a plate match before drawing a weapon or making a stop based solely on an automated alert?
If our town's cameras are ever paired with facial recognition or person-tracking software, who approved that, and was it disclosed before the hardware was installed?
Does our contract prohibit the vendor from enabling facial recognition or person-search features in the future without a new public vote, or does it leave that decision entirely to the company?
Who in our town is responsible for noticing if a software update adds a capability we never approved, and how would they find out?
This page will be updated as more factual sources become available.