Flock, a company known for automatic license plate readers and AI-powered cameras, relies on overseas workers from Upwork to help train its machine learning systems, according to material reviewed by 404 Media that was accidentally exposed by the company.
The disclosure matters because Flock’s products are not ordinary AI tools. Its cameras are used in thousands of communities in the US and are part of a surveillance system that can record details about vehicles, people, and movement. That makes the question of who reviews training footage, and where those reviewers are based, a central issue.
What the exposed material showed
Multiple tipsters directed 404 Media to an exposed online panel connected to Flock’s AI training work. The panel showed metrics tied to the process, including categories for “annotations completed” and “annotator tasks remaining in queue.”
In this context, annotations are the notes workers add when they review footage. Those notes help train machine learning algorithms. The tasks described in the exposed material included categorizing vehicle makes, colors, and types, transcribing license plates, and handling “audio tasks.”
The panel also listed people assigned to annotate Flock footage. Using those names, 404 Media found that some were located in the Philippines, based on LinkedIn and other online profiles. Many of the workers were employed through Upwork, according to the exposed material.
Upwork is a gig and freelance platform where companies can hire workers for services such as design, writing, and AI services, according to Upwork’s website.
Why Flock footage is unusually sensitive
Companies that build AI and machine learning systems often use overseas labor to label data, in part because it can be cheaper than hiring domestically. But Flock’s work sits in a more sensitive category because its cameras are designed to monitor public movement.
Flock cameras continuously scan the license plate, color, brand, and model of vehicles that pass by. Law enforcement can then search cameras nationwide to see where else a vehicle has appeared. Authorities typically search this data without a warrant, according to the source article.
Flock’s technology has become widespread in the US. Police use the cameras every day to investigate incidents such as carjackings, and local police have performed numerous searches for ICE in the system.
The American Civil Liberties Union and Electronic Frontier Foundation recently sued a city covered by nearly 500 Flock cameras. The source article connects that lawsuit to broader concerns about warrantless access to vehicle-location data.
What workers were asked to label
The exposed materials and public Flock presentations described several types of labeling work. Workers were instructed to categorize vehicles by make, color, and type, and to transcribe license plates.
The material also involved people and audio. Broadly, Flock uses AI or machine learning to automatically detect license plates, vehicles, and people, including clothing, from camera footage. A Flock patent also mentions cameras detecting “race.”
It is not clear what specific camera footage Flock’s AI workers are reviewing. However, screenshots in worker guides showed many images of vehicles with US plates, including plates from New York, Michigan, Florida, New Jersey, and California.
Other images contained road signs showing the footage was taken inside the US. One image included an advertisement for a specific law firm in Atlanta.
The worker guidance also covered audio. One slide instructed workers to listen to an audio clip all the way through before choosing from categories such as “car wreck,” “gunshot,” and “reckless driving.” Another slide connected tire screeching with someone “doing donuts.”
Flock has recently begun advertising a feature that will detect “screaming.” The source article says one slide explained that it can be difficult to distinguish between an adult and a child screaming, so workers were asked to mark their confidence with options such as “certain” and “uncertain.”
The core accountability question
The central issue is not only whether AI training is outsourced. It is whether people reviewing surveillance material may have access to footage of US residents, vehicles, roads, and audio captured by systems used by law enforcement.
The source article does not establish exactly which footage the workers reviewed. It does show that Flock training material included examples with US license plates, US road signs, and other indicators that the footage came from inside the United States.
That distinction is important. A generic AI training job might involve labeling everyday images or audio. Flock’s system, by contrast, is tied to a network that can help trace where vehicles have traveled and can detect people and other scene details.
For communities using Flock cameras, the reporting raises practical questions: who has access to the training pipeline, what footage is included, how worker access is controlled, and what safeguards apply when law-enforcement surveillance data is used to train AI systems.
What happened after Flock was contacted
After 404 Media contacted Flock for comment, the exposed online panel was no longer available. Flock then declined to comment.
That leaves the public with a limited but significant picture. The exposed material indicates that Flock used overseas Upwork workers for AI training tasks involving surveillance-related footage. It also shows that the company’s training materials directed workers on how to label vehicles, license plates, people, and audio events.
For an AI camera company operating across thousands of US communities, that is enough to make access, transparency, and oversight the main questions now facing the system.