US e-commerce giant Amazon.com Inc has issued a one-year temporary suspension on the police use of its facial recognition software, Rekognition. The company officially announced the moratorium on Wednesday, reasoning that there needs to be stronger regulations put in place for the continued use of the technology.
In a blog post on its website, Amazon stated that it is calling for governments to come up with new regulations for the ethical use of facial recognition software. It added that the moratorium should give Congress and lawmakers ample time to write and implement the new regulations.
Earlier in the week, House Democrats had introduced new legislation that would require law enforcement agencies to have a warrant before they could run facial recognition software on faces taken from an officer's body camera.
For the meantime, Amazon is allowing the limited use of its software for agencies that need it the most. This includes groups such as the International Center for Missing and Exploited Children.
Amazon's Rekognition software is one of the most advanced facial recognition software available for commercial use. The cloud-based platform is able to match faces with previously submitted photos. The technology can recognize other objects as well, including animals, furniture, and other scenery.
Since the launch of the service in 2016, Amazon has been pushing for the use of its technology for law enforcement. Since its adoption, some police departments have used Rekognition to identify perpetrators whose faces were captured in photos or on video using surveillance cameras.
Rekognition is then able to compare the faces to thousands of previously submitted mug shots, giving a similarity score that indicates how confident it is of a match. It is not yet clear how many police departments are actively using Amazon's technology as of today.
Amazon has received some criticism over the proposed accuracy of its software. In 2018, the American Civil Liberties Union (ACLU) had uploaded more than 25,000 mug shots to Amazon's cloud-based servers. The group then used Rekognition to identify photos of members of Congress. The experiment resulted in around 28 out of the 535 members of Congress being mistakenly matched to known criminals.
The exercise understandably got the attention of several members of Congress, who had then launched an inquiry into the use of the technology. Amazon released a statement following the experiment, stating that ACLU had intentionally lowered the minimum threshold for matching photos to around 80 percent.
Amazon explained that for a more accurate match, the threshold would have to be set higher. The company later stated that law enforcement agencies should set the threshold to 99 percent when conducting facial searches.
