Meta's ships facial recognition on smart glasses
Meta's smart glasses app contains a fully functional facial recognition pipeline, currently dormant for most users.
A reverse-engineering analysis of Meta's Stella app (version 273.0.0.21) found three on-device face-recognition models totaling ~100 MB: SCRFD (face detection), KPSAligner (alignment), and SFace (embedding generation). The pipeline produces 2048-dimensional biometric fingerprints and stores them in a cosine-similarity vector index in a local SQLite database under RLDrive, Meta's cross-device sync framework. The system writes unrecognized faces to a directory called NameTagsPending as both cropped image files and binary embeddings. A fully wired notification channel exists to alert users when a recognized face matches the index. The researcher confirmed the pipeline functions end-to-end by triggering recognition on test images, but found no evidence of active enrollment or server-side data flow on standard accounts. The user-facing UI elements (Connections card, profile destination screen) are either absent or hidden from the build.
What HN community is saying
Commenters expressed alarm at the capability's existence independent of current deployment status. A former Facebook researcher noted that facial recognition was heavily restricted during their tenure, with legal and FTC pressure forcing face removal as a core feature; they attributed the shift to either Zuckerberg's explicit mandate or changed regulatory conditions under the current administration. Multiple commenters cited a New York Times report quoting Meta's internal strategy document stating they would launch during a political environment where civil society opposition would be distracted. Workplace security policies banning smart glasses emerged as a practical response. One commenter noted that Wired reported Meta claimed to delete facial recognition data on 1 billion people in 2021, but questioned whether that deletion actually occurred.