Audits of Flock Safety's AI-powered license-plate reading cameras show significant error rates in identifying vehicles and generating police alerts.
These inaccuracies raise concerns about public safety and civil liberties, as false positives can lead law enforcement to stop innocent drivers based on flawed data.
An audit of the Los Angeles Police Department's system covering a two-month period in 2024 found a 32.3 percent error rate [1]. During that window, the cameras generated 161 false stolen-vehicle alerts [1]. The same system successfully resulted in the recovery of 337 stolen vehicles [1].
Other reports indicate even higher failure rates. In Roseville, California, 71 percent of the 1,427 incidents flagged by Flock cameras were wrong [6, 7]. Separate reporting suggests that Flock cameras may have misread over 70 percent of license plates [4].
Law enforcement agencies use these automated license-plate recognition systems to track suspects and recover property. However, the discrepancy in accuracy reports, ranging from roughly one-third to over 70 percent, highlights a lack of consistency in how the AI performs across different deployments [1, 5].
The technology is deployed in various U.S. cities to improve response times and solve crimes. Critics argue that the high rate of misidentification creates a public safety hazard by diverting police resources and risking unnecessary confrontations with the public [3, 5].
“Flock cameras may have misread over 70 percent of license plates.”
The gap between the marketed reliability of AI surveillance and the audited reality suggests a systemic risk in automated policing. When error rates reach 71 percent, the technology may function more as a source of noise than a reliable investigative tool, potentially undermining the legal justification for stops and searches based on these alerts.
