ROSEVILLE, Calif. — An analysis by the Roseville Police Department found that Flock Safety’s automated license plate recognition cameras incorrectly read license plates in 71% of stolen-vehicle and felony alerts reviewed by police, raising concerns about the reliability of AI-powered surveillance technology increasingly used by law enforcement.
Between 2023 and 2024, Flock sent 1,427 alerts to Roseville police after its cameras identified vehicles as stolen or potentially connected to felony crimes. Police analysis found that the company’s machine-learning software incorrectly read license plates in 71% of those alerts, according to reporting by Business Insider cited in a report by CNET.
The figure does not mean that Flock incorrectly read 71% of every license plate captured by its cameras. Rather, it reflects the error rate among the 1,427 stolen-vehicle and felony alerts reviewed by Roseville police — precisely the type of alerts that could potentially trigger a law enforcement response.
The findings raise concerns about the role of artificial intelligence in criminal investigations, particularly when officers rely on automated technology to identify vehicles potentially connected to criminal activity. An incorrect reading can result in an innocent vehicle being identified as one reported stolen or associated with a felony investigation.
Records reviewed by Business Insider showed that Roseville police repeatedly encountered problems with the system, including incorrectly read license plate characters and states, blurry images, missed vehicles and delayed alerts.
The report highlighted how even a single incorrect digit can have significant consequences. In one Roseville case, Flock repeatedly flagged a man’s vehicle as stolen or connected to a felony because the technology mistook a “9” on his license plate for an “8.”
The mistake happened at least six times, according to the reporting. The vehicle owner agreed to remove a license plate cover that officials thought might be contributing to the problem, but the incorrect alerts continued.
Such errors raise the possibility of unnecessary police encounters when automated surveillance technology incorrectly identifies an innocent driver’s vehicle as being associated with criminal activity.
Roseville police said officers are required to independently verify Flock alerts before taking enforcement action, providing a layer of human review between the automated identification and a police stop.
The department said it was not aware of any wrongful stops or arrests in Roseville resulting from the inaccurate alerts. However, cases elsewhere involving erroneous automated license plate reader information have raised concerns about innocent motorists being subjected to police encounters because of incorrect data.
Flock has challenged the implication that the Roseville findings represent the typical performance of its cameras.
The company says that under optimal conditions its cameras accurately read more than 96% of license plate characters. Flock has also pointed to what it described as Roseville’s unusual deployment of the cameras as a factor contributing to the city’s problems.
Roseville configured its cameras to photograph only the rear of vehicles in an effort to avoid collecting personally identifiable information, including drivers’ faces.
According to Flock, Roseville was also using older hardware, with cameras positioned higher and farther away from vehicles than the company typically recommends. The company said those conditions made Roseville’s deployment more challenging than its standard installations.
Camera placement, lighting conditions, dirty license plates and license plate covers can also affect automated plate-reading accuracy.
The dispute illustrates a broader challenge surrounding automated surveillance systems: The accuracy achieved under controlled or optimal conditions may differ significantly from performance when the technology is deployed on public streets.
Flock said the performance of Roseville’s cameras has significantly improved since the 2023-24 period examined by police.
Roseville, however, has not confirmed that claim. The department said its accuracy data for 2025 and 2026 are incomplete and that it has not conducted an analysis sufficient to determine how much performance has improved.
More recent records also indicate that problems have continued.
In April 2025, a Roseville police supervisor informed Flock that 16 of 17 license plates flagged during the previous week were incorrect. Some involved what the supervisor described as obvious misreads in which the cameras confused characters that did not closely resemble one another.
Police also documented instances in which Flock cameras failed to identify vehicles officers were actively trying to locate, including vehicles associated with investigations into mailbox burglaries, a deli break-in and reckless driving.
Flock apologized for errors identified by the department and proposed possible solutions, including repositioning cameras and implementing software updates intended to address blurry plates and overexposed images.
The Roseville Police Department has reported problems to Flock dozens of times and continues to meet with company representatives twice a month to discuss performance and other concerns.
Despite those problems, Roseville continues to use Flock’s technology.
The city began using Flock cameras in 2021 and has spent approximately $450,000 on the company’s products since 2020. Police officials say the technology has helped investigators solve crimes, including cases they believe otherwise might have remained unsolved.
Flock cameras, typically mounted on poles or street signs, use machine-learning technology to capture license plate information along with vehicle characteristics such as color, model, dents and bumper stickers.
Law enforcement agencies can use that information to search for vehicles believed to be connected to crimes, and in some circumstances officers can search across networks containing cameras operated by numerous agencies.
That capability has made automated license plate readers increasingly attractive to police departments, but it has also intensified concerns among privacy and civil liberties advocates over the scale of government vehicle tracking and the sharing of information among law enforcement agencies.
Flock says its cameras are operating in more than 6,000 communities nationwide and process more than 20 billion vehicle reads each month.
At the same time, dozens of communities have canceled, paused or suspended Flock contracts amid concerns about privacy, potential misuse of the technology and the sharing of license plate information with federal immigration authorities.
According to CNET, several cities, including Los Angeles, have ended contracts with Flock Safety or transitioned to other surveillance providers amid concerns over privacy and government surveillance.
The Roseville findings add another issue to that debate: whether automated surveillance technology is sufficiently accurate when its information could become part of a criminal investigation.
The findings have also drawn attention from lawmakers.
California Assemblymember Josh Lowenthal, who previously introduced legislation restricting the retention of automated license plate reader data, called for a statewide examination of the reliability of license plate reader systems following disclosure of the Roseville findings.
U.S. Sen. Alex Padilla has also raised concerns about automated license plate reader technology and the potential consequences of inaccurate or improperly shared data.
The controversy highlights the importance of safeguards separating an automated identification from police enforcement.
Roseville’s policy requiring officers to verify alerts before acting demonstrates one potential safeguard. But critics argue that relying on individual agencies to establish their own procedures leaves unresolved questions about statewide standards, transparency, auditing and accountability.
Although automated license plate recognition systems may allow police to search for vehicles more quickly than officers could manually, inaccurate readings can generate false investigative leads, increase officers’ workloads and potentially expose innocent people to unnecessary law enforcement scrutiny.
The Roseville experience also demonstrates the tension between privacy and technological performance. The city’s decision to position cameras in a way designed to avoid capturing people’s faces may have contributed, according to Flock, to poorer license plate recognition.
As more jurisdictions consider expanding automated license plate recognition systems, the Roseville analysis underscores the need to evaluate not only what the technology can accomplish but how reliably it performs under actual law enforcement conditions.
Supporters argue that the technology can improve public safety by helping officers locate suspects and solve crimes more quickly. Critics maintain that inaccurate identifications, widespread vehicle tracking and questions about access to collected data demonstrate the need for stronger oversight, transparency and accountability before such surveillance systems become further embedded in everyday policing.
To comment please go to one of these social media platforms:
Facebook: https://www.facebook.com/DavisVanguard
LinkedIn: https://www.linkedin.com/company/peoples-vanguard-of-davis-inc/
Instagram: https://www.instagram.com/davisvanguard/
X: https://x.com/davisvanguard
Bluesky: https://bsky.app/profile/vanguardnewsgroup.bsky.social
Subscribe to the Vanguard News letters. To make a tax-deductible donation, please visit davisvanguard.org/donate or give directly through ActBlue. Your support will ensure that the vital work of the Vanguard continues.
