How Accurate Are Flock's AI-Powered License-Reading Cameras?

A July-10 audit by the Los Angeles Police Department Office of the Inspector General found that the department’s ALPR cameras generated 161 false stolen-vehicle alerts in a two-month span, translating to an overall error rate of 32.3 percent. Each false alert resulted in a traffic stop of an innocent driver, illustrating a direct operational cost in officer time and public trust. The audit also recorded 337 alerts that correctly recovered stolen vehicles, showing the system’s mixed performance. This data provides a concrete baseline for engineers evaluating the risk-benefit of deploying similar cameras in other jurisdictions.

In Roseville, California, police records show 1,427 alerts over 2023-2024, with 71 percent of those stemming from misread license plates. The city’s unique deployment, cameras aimed only at vehicle backs, older hardware, higher mounting, and greater distance, exacerbated blurry images and partial plate captures. Additionally, license-plate frames that obscure character edges were cited as a source of confusion, especially between similar characters like “9” and “8.” The analysis indicates that deviations from Flock’s recommended hardware and placement can dramatically increase false-positive rates.

Flock’s public statements assert that, under optimal conditions, its cameras correctly read more than 96 percent of license-plate characters and that false alerts occur at less than one in a million when officers express doubt. The company also promotes a “suppression” mechanism allowing local agencies to mute alerts for plates known to be resolved, attempting to work around the static, comma-separated NCIC database. However, the suppression model relies on multiple agencies flagging bad entries, a process that is not automated and therefore may not keep pace with real-time errors, limiting its practical impact.

The operational fallout includes police officers pulling over innocent motorists, as seen in the LAPD and Toledo incidents, leading to injuries, job loss, and settlements such as a $35,000 payout. These outcomes force agencies to implement additional verification steps before acting on an alert, increasing response latency and staffing overhead. Engineers must therefore design systems that either reduce false alerts through better hardware and configuration or provide robust feedback loops to mitigate downstream legal and reputational risks.

For teams considering adoption, the cost includes not only the hardware purchase but also the need to follow Flock’s recommended camera placement, upgrade older units, and possibly modify local policies to require secondary confirmation before stops. Deployments that cannot meet these conditions will likely experience error rates similar to Roseville’s 71 percent, eroding the value of the ALPR investment. Consequently, a careful pilot that measures real-world false-positive rates against the claimed >96 percent accuracy is essential before scaling.