Facebook said Tuesday it would drastically scale back its facial recognition system and delete more than 1 billion users' facial templates after years of mounting criticism over the company's practices.
The announcement comes several months after a federal judge approved a settlement of a class action lawsuit in Illinois in which Facebook agreed to pay $650 million for allegedly using face-tagging and other biometric data without the permission of users.
Privacy advocates and lawyers for years have questioned Facebook's practice of scanning photos for any recognizable faces and have accused the company of misrepresenting the system to people using the social media platform.
Facebook said in its statement from Jerome Pesenti, a company vice president for artificial intelligence, that it would still use facial recognition technology for a "narrow set of use cases." Those would include helping people gain access to a locked account, verify their identity in financial products or unlock a personal device, he said.
But for other situations such as photos on social media, the company said it had re-evaluated whether the system was worth the trouble and the costs.
"We need to weigh the positive use cases for facial recognition against growing societal concerns, especially as regulators have yet to provide clear rules," Pesenti said.
"As part of this change, people who have opted in to our Face Recognition setting will no longer be automatically recognized in photos and videos, and we will delete the facial recognition template used to identify them," he said.
He said that more than a third of Facebook's daily active users had opted in to the facial recognition setting and were able to be recognized. And he said some people put a high value on the system, including a feature that generates descriptions of images for people who are blind and visually impaired.
But he said there was too much uncertainty around regulation and privacy concerns.
"Amid this ongoing uncertainty, we believe that limiting the use of facial recognition to a narrow set of use cases is appropriate," he said.