How Delhi Police's Facial Recognition System Failed At CJP Protest | Open Intel

Our Take

Credit to India Today Global's Open Intel series for doing what most outlets won't: actually explaining the math behind facial recognition instead of just marveling at it. The core point in this episode is the one we keep hammering here at DeFlock — the machine never says 'yes.' It spits out a similarity score, and a human being decides where the cutoff is. Lower that threshold to catch more people, and you catch more wrong people too. That's not a bug in the software; it's the entire business model.

The Delhi Police case is a preview of what's already happening stateside with license plate readers. Flock Safety's network doesn't identify faces, but it runs the same playbook: an algorithm generates a 'match' or a 'vehicle of interest' alert, and an officer treats it as probable cause instead of a starting point. The ACLU's tally of wrongful arrests tied to facial recognition in the US — at least fourteen and counting — should be read alongside the growing pile of ALPR misreads that have sent guns drawn on innocent drivers. Different sensor, same failure mode: opaque confidence scores get laundered into false certainty by the time they reach a badge and a holster.

What ties Delhi's protest surveillance to Flock's 40,000+ camera network in American towns is the underlying logic — cast a wide net, trust the algorithm, sort out the mistakes later, and hope nobody asks who set the threshold. If you want to see how much of that net is already strung up where you live, check our camera map, and if you're ready to push back on it locally, our take-action page has the tools to start.