The ACLU may be on to something: According to a blog post published today, the ACLU used Amazon’s law enforcement facial recognition tool “Rekognition” on all the members of Congress, and turned up 28 (false) matches to people who have previously been arrested for a crime. The ACLU tested official Congressional portraits against a database of 25,000 publically available mugshots using Rekognition’s default match settings.

Amazon Rekognition has been decried by almost 70 civil and digital rights groups, including the EFF, as well as Amazon employees and shareholders, on account of concerns about the tool’s potential to track innocent people and falsely identify them as having committed a crime — precisely what happened in the ACLU’s test. The tool works by analyzing the arrangement of major facial features, comparing that arrangement to a database of mugshots, and spitting out a likelihood of a match as a percent, with positive matches shown in green text.

The government has done nothing to reel in the use of facial recognition in police departments, as places like Orlando, Florida and Washington County, Oregon have already piloted Amazon Rekognition. The appeal of Rekognition is that it’s incredibly cheap to use, and the ACLU claims its test cost just $12.33 to run. It appears you get what you pay for.

Amazon Rekognition has been criticized on grounds of enabling the heightened surveillance of populations vulnerable to over policing, like people of color and immigrants. Eleven of the falsely identified Congresspeople in the ACLU’s test are people of color — 40 percent of the total false matches — even though people of color make up less than 30 percent of Congress. One of the positive matches in the ACLU’s test was representative John Lewis, who was arrested in the 1960s during civil rights protests, but the blog post implied that his actual mugshots were not included in their database.