What Meta’s Hidden Face Recognition System Is
Meta’s smart glasses face recognition system is a hidden set of machine-learning models and databases inside the Meta AI companion app that can detect faces, convert them into biometric fingerprints, and match them against stored faceprints on a user’s device without any visible feature being offered to the public. Recent investigations into the Meta AI app, which supports Ray-Ban and Oakley smart glasses, show unreleased code for a feature internally called NameTag, later rebranded in-app as “Connections.” The app has been downloaded more than 50 million times, which means this dormant infrastructure already sits on millions of phones. While Meta insists it is only “exploring” such capabilities and has not shipped them to consumers, the level of technical completeness suggests a system far beyond early experimentation and much closer to a ready-to-launch product.
Inside the Hidden Pipeline: From Faces to Biometric Fingerprints
Reverse engineering of the Meta AI app (internally named Stella) reveals a full recognition pipeline designed for Meta smart glasses face recognition, not a simple camera helper. According to Gadget Review, the app ships three ExecuTorch models that work together: SCRFD detects faces, KPSAligner aligns them, and a scaled-up SFace variant turns each face into a 2048‑dimensional biometric fingerprint. An SQLite vector database supports similarity matching, while persistent storage keeps cropped images and binary embedding files of unknown faces in a private directory that survives reboots. During tests by researchers at Buchodi, the system identified a portrait of Michel Foucault once his biometric data was preloaded, demonstrating an end‑to‑end match. Quote-worthy, Gadget Review notes that this “technical sophistication rivals dedicated surveillance systems,” even though regular users never see any visible face recognition feature or control in the app.
Meta’s Public Story vs. the Code on Your Phone
The gap between what Meta says and what its apps contain is driving many smart glasses privacy concerns. WIRED-reviewed code in the live Meta AI companion app shows components of NameTag present as early as January, including three on-device AI models for detection, cropping, and biometric conversion. A later build surfaces user-facing text for a feature called “Connections,” inviting people to “remember the people you met,” plus a hidden notification string saying “Person recognized.” Yet none of this appears in the live interface. Meta spokesperson Ryan Daniels told WIRED these findings are “merely evidence” that the company is exploring options and that “nothing has shipped to consumers and no final decision has been made.” The company also says it will not build a central face database and promises any rollout would come with full transparency, but the dormant code already distributed to millions suggests a more advanced, quietly staged deployment.
Biometric Data Privacy and What Users Don’t Realize They’re Allowing
The hidden face recognition system raises serious biometric data privacy questions, because it builds an on-device archive of faces and embeddings before users understand what they are agreeing to. The Meta AI app’s disclosures already admit that data from AI glasses may be stored on both local devices and remote servers, yet there is no clear, dedicated consent flow for creating 2048‑dimensional face fingerprints or keeping unknown faces as “pending identification” entries. In an Android Authority poll about facial recognition in smart glasses, 71% of 84 respondents chose “Oh hell no!”, hinting that most people would not lightly accept this level of tracking. Without prominent controls or clear explanations, people who only wanted AI-assisted eyewear may find that their everyday interactions are being quietly converted into long-lived biometric records, with no easy way to understand, audit, or erase what has been captured.
Regulatory Risks and the Future of Smart Glasses Privacy
Shipping a hidden face recognition system in a mass-market companion app puts Meta under renewed scrutiny from regulators already wary of biometric surveillance. Meta shut down its earlier Facebook face recognition system in 2021, citing concerns over how sensitive this technology is, yet the company is now distributing sophisticated recognition tools tied to smart glasses without a launched feature or clear opt-in. The company insists it is not building a central face database, but on-device databases still create risks: phones can be compromised, data can be synced, and features can be switched on with a software flag. For users, the discovery underscores how little control they have over what their wearables are technically capable of. Until face recognition in smart glasses is treated as an explicit, high-stakes consent decision, hidden pipelines like this will remain a flashpoint in the fight over biometric surveillance.
