Flock Employs International Freelancers to Develop Its Surveillance AI

Flock, the automated license plate reader and AI-driven camera firm, employs remote workers from Upwork to enhance its machine learning algorithms. Training materials indicate that these workers are instructed on how to examine and classify footage, including images of individuals and vehicles within the United States, as noted in documents reviewed by 404 Media that were inadvertently revealed by the company.

These revelations raise concerns about who has access to footage obtained by Flock’s surveillance cameras and the locations of the personnel analyzing this footage. Flock’s technology has become widespread in the US, with its cameras installed in numerous communities utilized by law enforcement to investigate incidents such as carjackings. Local police have also conducted various lookups for ICE in the system.

Companies leveraging AI or machine learning frequently rely on overseas workers to train their systems, primarily due to lower labor costs compared to domestic hiring. However, Flock’s business—developing a surveillance infrastructure that consistently tracks the movements of US residents—suggests that the footage may carry higher sensitivity than other AI training assignments.

Flock’s cameras repeatedly capture the license plate, color, make, and model of all vehicles that pass by. Authorities can then search the cameras nationwide to track where else a vehicle has been spotted. This data is typically accessed without a warrant, prompting the American Civil Liberties Union and Electronic Frontier Foundation to recently file a lawsuit against a city saturated with nearly 500 Flock cameras.

In general, Flock employs AI or machine learning to automatically recognize license plates, vehicles, and individuals, along with details such as their clothing. A patent from Flock also mentions the capability of cameras to detect “race.”

Several anonymous sources directed 404 Media to a publicly accessible online panel that displayed various metrics related to Flock’s AI training.

This panel included data on “annotations completed” and “annotator tasks remaining in queue,” with annotations referring to the notes added by workers to the analyzed footage to facilitate AI training. Tasks encompass categorizing vehicle makes, colors, and types, transcribing license plates, and performing “audio tasks.” Flock recently introduced a feature that claims to detect “screaming.” The panel indicated that workers occasionally completed thousands of annotations over two-day spans.

The exposed panel also listed individuals responsible for annotating Flock’s footage. Through their LinkedIn and other online profiles, 404 Media discovered some were based in the Philippines.

Numerous individuals were reportedly hired through Upwork, as indicated in the exposed documents. Upwork serves as a platform for gig and freelance work, allowing companies to hire designers and writers or procure “AI services,” according to Upwork’s official site.

The sources also highlighted several publicly available Flock presentations that elaborated on how workers were to categorize the footage. It remains unclear which specific camera footage Flock’s AI workers are assessing. However, screenshots from the worker guides feature multiple images of vehicles with US license plates from states such as New York, Michigan, Florida, New Jersey, and California. Other images show road signs clearly indicating that the footage originates from within the US, including one featuring an advertisement for a specific law firm in Atlanta.