This City’s AI License Plate Cameras Were Wrong 7 Out Of 10 Times

Internal emails have revealed that Flock Safety, a company providing automated license plate readers, worked to improve camera accuracy after a California city, Roseville, documented widespread misreads. The city found that 71 percent of Flock alerts, out of 1,427 generated during 2023 and 2024, stemmed from incorrect license plate reads. The cameras were found to confuse similar-looking characters, capture blurry images, or miss vehicles altogether. In one instance, a driver’s plate was repeatedly mistaken for that of a stolen vehicle due to the software misreading a “9” as an “8”. Despite the high error rate, the city continues to use the system, citing its benefits in solving crimes. Flock has acknowledged the issues, stating that Roseville’s deployment was unusual, with rear-facing cameras mounted higher and farther from traffic than recommended, and using older hardware. The company has been working to improve the system, including developing tools to automatically detect missed vehicles. The incident highlights the importance of treating AI-generated alerts as investigative leads rather than unquestionable facts, and the need for officers to independently verify information before taking action.