The American Civil Liberties Union tested out Amazon's facial recognition technology, and they say it falsely matched 28 current members of Congress with images of people who had been charged with crimes.
ACLU of Northern California tech and civil liberties attorney Jacob Snow says the testing also found that Amazon's software — used by some police departments — disproportionately misidentified black and Latino people.
In their testing, the ACLU used Amazon Rekognition to compare the photos of all federal lawmakers to a database of 25,000 mugshots.
"We're calling for immediate action by Congress to put the brakes on this technology with a moratorium on its use by government," Snow said. "We need to have a discussion about whether or not face recognition is safe for use at all."
Amazon Web Services spokeswoman Kathleen Rohde said in a statement that the civil liberty group's test results came from the fact that they used an 80 percent confidence threshold, which is lower than what they would recommend for law enforcement purposes.
"While 80 percent confidence is an acceptable threshold for photos of hot dogs, chairs, animals, or other social media use cases, it wouldn’t be appropriate for identifying individuals with a reasonable level of certainty," she said. "When using facial recognition for law enforcement activities, we guide customers to set a threshold of at least 95 percent or higher."
Three of the misidentified lawmakers, including East Bay congressman Mark DeSaulnier, penned a letter to Amazon CEO Jeff Bezos on Thursday questioning how the company has tested the technology for accuracy and bias. The lawmakers also demanded a list of government agencies that use Rekognition.
DeSaulnier told KQED on Thursday that he thinks the technology just isn't ready for use by law enforcement.
"Our justice system is not perfect, and too many people have convicted crimes that they're innocent of," DeSaulnier said. "So I just think we need to go slowly with this technology until we're sure that it actually works."
Amazon Rekognition is currently being used by the Washington County Sheriff's Office in Oregon. According to a customer testimonial posted on Amazon's website, the agency has used the technology to expedite its suspect identification process.
"Within one week of going live, the solution built on Rekognition identified a suspect for a cold case that led to an arrest through due process," said Chris Adzima, a senior information systems analyst for Washington County Sheriff’s Office, in the testimonial.
Snow says the ACLU's testing mimicked the process used by the Washington County Sheriff's Office's.
"This test is in parallel with those deployments," he said.
Snow says Amazon's facial recognition technology has also been used by the Orlando Police Department. An ACLU investigation released in May found that Amazon aggressively marketed Rekognition to the OPD as a tool for surveillance.
"They were saying that the Orlando police could use it for real-time tracking, and they could use it to track people as they move around the city," Snow said.
"Technology should not be used for mass surveillance, and it shouldn't be used to track people as they go about their daily lives," he said.
KQED's Ted Goldberg contributed to this report.