Code reviewed by WIRED uncovered an unreleased face-recognition system embedded in Meta\u2019s smart glasses platform. It\u2019s designed to identify people via biometric data stored on users\u2019 phones.<\/p>\n\n\n\n Meta has quietly embedded face-recognition<\/a> technology for its smart glasses<\/a> into an app downloaded to millions of phones, according to a WIRED analysis of the company’s software.<\/p>\n\n\n\n Code discreetly added to Meta\u2019s AI app over multiple updates this year shows that the feature, internally called \u201cNameTag,\u201d identifies people captured by the glasses\u2019 camera<\/a> and, when activated, alerts the wearer when it recognizes someone.<\/p>\n\n\n\n The discovery of NameTag in the live Meta AI app shows that Meta had begun shipping face-recognition code to users’ phones while publicly describing it as something the company was still \u201cthinking through.\u201d In April, Meta said if it were to utilize face recognition, it wouldn’t be rolled out without first taking “a very thoughtful approach.” But WIRED found that as early as January, core components of the system had been integrated into software distributed to millions of people.<\/p>\n\n\n\n Though not yet enabled, NameTag sits inside a Meta AI companion app that’s been downloaded over 50 million times and is necessary for use of key features of its smart glasses, including Ray-Ban and Oakley models. If activated, it will transform faces captured by Meta’s glasses into unique biometric signatures, commonly known as faceprints, and check each one against faceprints stored on the user\u2019s phone\u2014a database that\u2019s currently configured to receive updates from Meta. Recognized faces will trigger notifications, while the rest are cropped, indexed, and saved to a folder marked \u201cpending.\u201d<\/p>\n\n\n\n NameTag would revive a type of technology Meta said it had sunsetted in 2021, when the company announced it would delete more than a billion faceprints belonging to Facebook users following years of controversy over its photo-tagging system. Meta ultimately paid $650 million to settle a class-action lawsuit brought by Illinois users and, in 2024, agreed to a separate $1.4 billion settlement with Texas over allegations it had unlawfully collected biometric data from users.<\/p>\n\n\n\n Its renewed efforts arrive amid mounting opposition to consumer-level face recognition, which privacy advocates argue will give anyone from stalkers to immigration agents easy access to a dangerous technology. Internal Meta documents published by The New York Times<\/a> in February showed the company had planned to roll out the feature during a \u201cdynamic political environment,\u201d when Meta believed its biggest critics would be preoccupied.<\/p>\n\n\n\n Three AI models powering NameTag have already been deployed from Meta’s servers and now reside on its customers’ phones, according to WIRED\u2019s analysis, which was independently reproduced by outside experts. One model detects faces, one crops them, and a third encodes them into biometric data.<\/p>\n\n\n\n Only traces of the user interface are currently present, hinting at how the feature may ultimately work. A May version of the app rebrands the feature for users as \u201cConnections,\u201d inviting them to \u201cremember the people you met.\u201d It remains unclear whose faces will be included in the system’s recognition database, how those profiles are created, or how many people could ultimately be identifiable through it.<\/p>\n\n\n\n WIRED shared its findings with two outside security researchers who separately examined the app and reproduced key aspects of the analysis: Cooper Quintin, a security researcher and senior public interest technologist with the nonprofit Electronic Frontier Foundation\u2019s Threat Lab, and an independent security and privacy researcher who goes by the pseudonym Buchodi and has spent more than a decade reverse engineering consumer software and surveillance technologies.<\/p>\n\n\n\n \u201cThe feature is not yet exposed to consumers but seems nearly ready to go,\u201d says Quintin. \u201cDespite the billions of reasons not to, Meta seems to have created the capacity to turn their customers into a distributed surveillance machine.\u201d<\/p>\n\n\n\n Buchodi ran additional tests on the recognition pipeline. (Read their technical analysis here<\/a>.) To see if the matching system worked, Buchodi added a single faceprint to the app\u2019s gallery, taken from the deceased French philosopher Michel Foucault. After triggering NameTag with Foucault\u2019s image, the app produced a notification: \u201cPerson recognized.\u201d<\/p>\n\n\n\n \u201cThe main components of a face-recognition feature are already in Meta’s companion app,\u201d Buchodi says. \u201cNot many pieces stand between this and a working feature.\u201d<\/p>\n\n\n\n![\"Image](\"https:\/\/media.wired.com\/photos\/6a1f7cd62dd56ccdeabd0786\/1:1\/w_2560%2Cc_limit\/security_meta_final_@.jpg\")