Amazon's Facial Recognition Software Misidentifies Lawmakers As Possible Criminals

Our Take

Credit to CBS Miami for keeping this story in circulation, because it never stopped being relevant. When the ACLU ran Amazon's Rekognition against mugshot databases back in 2018, it didn't just produce a handful of glitchy errors — it flagged nearly 30 sitting members of Congress as people who'd been arrested, with the false matches skewing heavily toward lawmakers of color. That's not a rounding error. That's a pattern baked into the technology, and it's the same pattern that shows up whenever automated systems are asked to make judgment calls about who looks like a criminal.

We cover license plate readers, not faces, but the underlying problem is identical: a private company sells a black-box matching algorithm to government agencies, promises accuracy, and then the burden of proving the system screwed up falls on the person wrongly flagged — not the vendor. Flock Safety's ALPR network runs on the same trust-us model. There's no independent audit requirement, no public error rate, no meaningful recourse when a camera misreads a plate and sends police to the wrong driveway with guns drawn. We've documented exactly that kind of incident before, and it keeps happening because nobody with power is forced to answer for it.

The lesson from Amazon's 2018 embarrassment is that even well-funded corporate tech fails constantly at exactly the tasks it's being sold to do for policing, and lawmakers only paid attention because it happened to them. Everyone else is stuck hoping they never become a false positive. If you want to see how far this camera network has spread near you, check the map, and if you'd rather do something about it, our take-action page has next steps.