What Meta’s Hidden ‘NameTag’ System Is and Why It Matters
Meta’s hidden smart glasses facial recognition system, internally called NameTag, is a pipeline of on-device machine learning models that detect faces, generate biometric fingerprints, store them locally, and later match them to identify people through Ray-Ban smart glasses without clear, informed user consent or transparent controls. Researchers who reverse engineered the Meta AI companion app found three models working together to support Meta facial recognition: one to spot faces in images, another to align them, and a third to generate 2,048‑dimensional embeddings that function as biometric IDs. An SQLite vector database and persistent storage preserved known and unknown faces, building a quiet “faces pending identification” archive. Although NameTag never appeared as a visible feature, its dormant code, buried in an app installed tens of millions of times, turns a consumer wearable into a potential smart glasses surveillance platform.
How Face Recognition Code Reached Millions of Smart Glasses Users
The Meta AI companion app, required for several Ray-Ban smart glasses products, has reportedly been installed more than 50 million times, giving NameTag silent access to a huge user base. According to researchers cited by multiple investigations, Meta shipped three ExecuTorch models—SCRFD for face detection, KPSAligner for keypoint positioning, and a scaled-up SFace variant for embedding generation—inside the app. The system could crop faces from smart glasses photos, encode them into high-dimensional vectors, and store both images and embeddings in a private directory that survived reboots. Hardcoded strings for a “Connections” feature promised to “remember the people you met,” along with a notification reading “Person recognized,” yet these options never surfaced in the interface. The disconnect between a dormant back end and a hidden front end shows Meta deployed face recognition code long before any open discussion of consent, controls, or biometric data privacy.
Exposure, Fast Deletion, and Meta’s Shifting Explanation
WIRED’s investigation brought the hidden NameTag system to light, triggering a rapid response from Meta. Within 24 hours of the exposé, Meta pushed an update that removed face recognition libraries, alert workflows, and the storage locations for biometric data, leaving only scattered debug fragments. GadgetReview reports that NameTag had been technically close to launch despite being inaccessible, with tests even recognizing a portrait of Michel Foucault after his biometric template was preloaded. Meta, however, frames the code as internal experimentation. A company spokesperson, Ryan Daniels, stated that the code reflects testing rather than a planned consumer feature and said Meta is not building a centralized facial recognition database. At the same time, executives have criticized reporting as “dishonest,” a tension that fuels doubts about how far the system progressed and what, if anything, was synced beyond users’ phones.
Privacy, Consent, and the New Face of Wearable Surveillance
The NameTag episode highlights how wearables can normalize smart glasses surveillance even when features remain hidden. Meta previously announced the shutdown of Facebook’s facial recognition system and deletion of more than a billion faceprints after facing lawsuits and regulatory pressure, yet the new code shows continued interest in biometric identification. While Meta emphasizes that NameTag ran on-device and that it is not building a central face database, the system’s ability to identify strangers, store unknown faces “for future processing,” and expand a personal gallery of passersby raises deep consent and biometric data privacy concerns. People captured by Ray-Ban smart glasses have no straightforward way to opt out or verify deletion of their biometric signatures. The rapid removal of the face recognition code after exposure also underscores a broader accountability gap: powerful surveillance-grade infrastructure can be shipped first and explained—or erased—only once discovered.
