Amazon Facial Recognition Software Identifies 28 Members of Congress as Criminal Suspects

Facial recognition software sold by Amazon allegedly misidentified 28 lawmakers as being criminal suspects.

The software was being used by the ACLU to test Amazon’s claim that Amazon Rekognition provides extremely accurate facial analysis through photos and video. All members of Congress were put through the software’s databases that were built from thousands of public arrest records.

The highest rate of misidentification belonged to people of color, with 39% of the members being misidentified. The ACLU used the program’s default match settings, but in a response, Amazon says that the program should have been put at 95% match instead of 80%, the default. They also added that they guide law enforcement agencies that want to use the software towards the 95% threshold.

Source: Facial Recognition Software Wrongly Identifies 28 Lawmakers As Crime Suspects : NPR

Leave a Reply

%d bloggers like this: