We Tested Amazon's Facial Recognition Software

Our Take

Credit to the ACLU for doing what Amazon wouldn't: putting Rekognition through a real-world stress test. The result — 28 sitting members of Congress falsely matched to mugshots — isn't a bug, it's the whole business model. Face surveillance and ALPR networks like Flock both run on the same premise: scan everyone, sort them against a database, and trust the machine when it says "match." The ACLU's experiment shows exactly how shaky that trust is, especially for people of color and anyone unlucky enough to share a passing resemblance with someone in a police database.

This matters for the same reason DeFlock exists. Once a city buys into automated identification — whether it's a face or a plate — the errors get treated as probable cause, not glitches. Nobody voted on the accuracy threshold, nobody audits the false-positive rate in public, and the vendor keeps selling the next contract on the promise that this time it's better. We've heard that pitch before, from Flock's own marketing.

If a private company's software can mistake your senator for a criminal, don't assume the cameras logging your car's plate every day are somehow more reliable or more accountable. Check our map to see what's already watching your neighborhood, and use our take-action resources to push back before mass identification becomes the default in your town too.