Meta strips NameTag face-recognition code from Ray-Ban app

WIRED's follow-up on Meta's NameTag is a small master class in how 'the feature does not exist' can be technically defensible and functionally hollow at the same time. According to WIRED's reporting, code for a face-recognition system called NameTag was sitting inside the Meta AI companion app for Ray-Ban and Oakley smart glasses, an app installed on more than 50 million phones. One day after the story ran, Meta stripped nearly all of that code out.

The mechanics are the reason this matters. What WIRED found was not a stray flag, it was three AI models wired together: one to detect a face, one to crop and reposition it, and one to encode it as biometric data, with faces the system failed to recognise cropped, indexed, and stored locally for future processing. Meta vice president of communications Andy Stone told WIRED the feature is 'purely exploratory,' adding that no final decision has been made. CTO Andrew Bosworth called the reporting 'incredibly misleading' and 'absolutely dishonest.' Then the code came out of the app.

The line that will follow Meta around comes from internal material WIRED describes. The company was reportedly interested in launching NameTag during a 'dynamic political environment' in the US because 'civil society groups that we would expect to attack us would have their resources focused on other concerns.' A coalition of 75 organisations led by the ACLU had already written to Mark Zuckerberg demanding the company 'immediately halt and publicly disavow' the feature, under a campaign called Eyewear, Not Spyware.

The honest caveat is that the retrieved reporting does not show NameTag was ever switched on for end users, and Stone's 'exploratory' framing may be accurate in a narrow engineering sense. What the reporting does not give you is a clean answer on whether any biometric data was collected during internal testing, whether functionally similar code still lives elsewhere in Meta's stack, or under what name the same pipeline might return.

The forward read is less about NameTag itself than the norm it exposes. Once functional biometric code has shipped to 50 million devices, regulators, class-action lawyers and competitors from Apple to Snap get to treat 'we hadn't decided yet' as a very expensive answer. For anyone building on-device AI, the takeaway is the one Meta just paid for: dormant code inside a shipped app is not a private thought.