Just when I thought I couldn’t have a lower opinion of flock.
What does it take to become the most successful AI surveillance company in 2025? If you’re anything like Flock, the startup selling automatic license plate readers and facial recognition tech to cops, you don’t really need much AI at all — just an army of sweatshop workers in the global south.
Bombshell new reporting from 404 Media found that Flock, which has its cameras in thousands of US communities, has been outsourcing its AI to gig workers located in the Philippines.
After accessing a cache of exposed data, 404 found documents related to annotating Flock footage, a process sometimes called “AI training.” Workers were tasked with jobs include categorizing vehicles by color, make, and model, transcribing license plates, and labeling various audio clips from car wrecks.
In US towns and cities, Flock cameras maintained by local businesses and municipal agencies form centralized surveillance networks for local police. They constantly scan for car license plates, as well as pedestrians, who are categorized based on their clothing, and possibly by factors like gender and race.
Semi-related: https://deflock.me/
Funny thing is that ANPR is a solved issue that you can literally run on-device with minimal training. As in, you can literally run it on a $30 WiFi camera. With existing open models.
Facial recognition is a bit more tricky but there are open AI models that translate facial data to what is essentially a hash that can be compared to other faces with high precision, and that too can run on the same hardware.
Hell, my cheap $100 smart doorbell has built in facial recognition that doesn’t require any cloud connection or such. All on-device.
Benn Jordan did exactly that
Okay I suspected these shitcams to be bad… but not this bad. Wow.