Meta shuts down its face recognition system

Meta announced it would shut down Facebook’s facial recognition system and delete the individual face templates — the numerical representations of over a billion users’ faces — used to power it. The system had underpinned automatic tagging suggestions in photos and videos, a “Suggested Tags” prompt that recognised a user’s friends in newly uploaded images, and a feature that recognised users appearing in their own “Memories.” Meta said more than a third of daily active Facebook users had the setting turned on.

The company framed the decision as one of several changes made by its new “Meta” corporate structure and said it needed “to weigh the positive use cases for facial recognition against growing societal concerns, especially as regulators have yet to provide clear rules.” Facial recognition had been a persistent source of legal exposure for the company: it paid $650 million in 2020 to settle an Illinois biometric-privacy class action over the same tagging feature, one of the largest privacy settlements to that point, and regulators in the EU and several US states had moved to restrict biometric identification more broadly.

Meta was explicit that the retreat was partial, not principled: it said it would continue exploring narrower, on-device applications of facial recognition, such as verifying identity for account recovery or unlocking a device, that did not require storing a face template on Meta’s servers. Automatic Alt Text, an accessibility feature describing photo contents to blind and visually impaired users, kept detecting that a person was present in an image but lost the ability to say who.

The shutdown was one of the largest voluntary withdrawals of a deployed AI capability by a major platform to that point, and it came without a court order or new statute compelling it — a decision made in anticipation of regulation that had not yet arrived, rather than in response to one that had.