What Meta’s Hidden Face Recognition System Is
Meta’s hidden smart glasses face recognition system is a software pipeline in its companion app that quietly turns camera footage into biometric identifiers, compares them against a local database of faceprints, and signals a match to the wearer, all without clear disclosure or user-facing controls. At its core, this is facial recognition smart glasses infrastructure: three machine-learning models detect faces, align them, and convert them into 2,048‑dimensional fingerprints stored on the phone. Researchers found that the app, internally called Stella, includes an SQLite vector database for similarity search and persistent storage for both recognized and unknown faces. Hardcoded strings for a “Connections” feature promise to “remember the people you met,” suggesting Meta intended a social memory tool. Yet none of this appears in the consumer interface, meaning millions installed a Meta privacy surveillance system they could not see or configure.
Discovery of the ‘NameTag’ Code After Meta’s Public FRT Retreat
The hidden system, reportedly codenamed NameTag, was uncovered through reverse engineering and static analysis years after Meta announced in 2021 that it was shutting down Facebook’s facial recognition and deleting over a billion faceprints. WIRED, Buchodi researchers, and EFF’s Threat Lab independently confirmed that the Meta AI glasses companion app contains active facial recognition code, even if the feature is not exposed to users. During testing, researchers pre-loaded a face template—famously a portrait of Michel Foucault—and the glasses later identified it on sight, proving end‑to‑end functionality. According to EFF, the app could convert every face in the glasses’ field of view into numerical biometric data for biometric data tracking against a local database. Gizmochina reports that this code has appeared in multiple app updates since January 2024 and that the app has been installed more than 50 million times.
How the Biometric Pipeline Works on the User’s Device
Behind the scenes, Meta’s smart glasses face recognition runs a sophisticated on-device pipeline. Meta ships three ExecuTorch models: SCRFD for detecting faces in the camera feed, KPSAligner for aligning facial landmarks, and a scaled-up SFace variant to generate 2,048-number embeddings that uniquely represent each face’s geometry. These embeddings are stored in an SQLite vector database designed for similarity matching. When the wearer encounters a person, the system converts their image into an embedding and compares it against stored faceprints on the phone. If it finds a close match, a hidden notification system can announce “Person recognized” and deep-link to a profile page that does not yet exist in the public interface. Unknown faces are saved as cropped images plus binary embedding files in a private directory that persists across reboots, creating a “faces pending identification” cache without explicit consent.
Privacy, Consent, and Meta’s Surveillance Track Record
The discovery has reignited concerns about Meta privacy surveillance and the risk of turning smart glasses into distributed surveillance machines. Even though Meta says NameTag reflects internal experimentation, privacy advocates point to the company’s history. Meta previously paid USD 650 million (approx. RM3,000,000,000) to settle a BIPA lawsuit over mass facial recognition on photos and later faced a USD 1.4 billion (approx. RM6,400,000,000) biometric case in Texas before shutting down Facebook’s face system. EFF warns that the new glasses pipeline could log and store faceprints for every stranger in view, enabling pervasive biometric data tracking without those people’s awareness. Internal Meta documents, cited by EFF, describe ambitions to launch facial recognition during periods when civil society critics would be distracted, suggesting a deliberate strategy to limit scrutiny even as regulators and the public debate the ethics of always-on smart glasses face recognition.
Public Outcry, Meta’s Response, and What Comes Next
Once WIRED exposed the embedded NameTag system, Meta moved fast to contain the backlash. Company spokesperson Ryan Daniels said the code represented exploration, not a planned product, and emphasized that Meta was not building a central face database. Yet actions told a different story. According to EFF, less than 48 hours after public reporting, Meta pushed a June 5 app update that removed the facial recognition technology, the “Person recognized” notification hooks, and the machine learning models and databases needed for on-device identification. EFF calls this a “quiet acquiescence,” noting that Meta has refused to say whether NameTag might return or what happened to biometric data captured during internal tests. The episode highlights how facial recognition smart glasses can shift from dormant code to active surveillance overnight and underscores the need for stronger privacy laws and transparent consent frameworks before such features ever reach consumers.
