Face recognition and police, a bad match

Our Take

Credit to the ACLU of Northern California for keeping this receipt in circulation. Their 2018 test of Amazon's Rekognition software — which misidentified real people, including sitting lawmakers, as criminal suspects — became one of the clearest public demonstrations that automated identification tools aren't the neutral, infallible systems vendors market them as. A corporate promise to stop selling to police, made under pressure, is not a policy. It's a PR move that can be reversed or quietly ignored the moment public attention moves elsewhere.

That's the same pattern we track with Flock Safety and ALPR networks. Both technologies rely on pattern-matching systems trained and deployed without meaningful independent audits, both get sold to police departments as objective tools, and both have a documented history of getting it wrong in ways that land hardest on Black and brown communities. A license plate reader that misreads a plate can trigger the same kind of wrongful stop or worse that a bad facial match can. The lesson from Amazon's 2018 failure isn't just about face recognition — it's about what happens when we let private companies define the rules for surveillance tech they profit from selling to law enforcement.

If your town has cameras like these watching every car that passes, you deserve to know where. Check our camera map and see what accountability tools exist in our take-action resources.