Reverse-engineering reveals how Flock’s AI search works

Wired reporters have reverse engineered part of Flock’s software interface, offering a closer look at the tools available to police officers. Flock operates a network of license plate readers and cameras, with about 120,000 installed across the US.

The interface includes searches for vehicle types, colors, and features. A “free form” query can look for almost anything passing through a video frame, although it requires a full video camera rather than a license plate reader.

Wired also found an “AI-powered watch list” tool. It allows part of a city to be placed under continuous video surveillance until someone matching a description, such as “person wearing scrubs,” is spotted.

Flock warns that its AI can make mistakes, but the software does not provide a clear way to identify inaccurate results or measure their accuracy. Officers can rank results by relevance and accuracy, and that feedback is used to refine the results page. Some terms related to race, ethnicity, and religion are blocked.

Before accessing the software, users agree not to use it in illegal, harmful, or abusive ways. The agreement also says that all queries and metadata are recorded for security and auditing. Flock has limited how long data is retained, but continues to face questions from state authorities and the public about legitimate use.