Security researcher Buchodi published analysis on June 4, 2026, revealing that Meta's Stella companion app for Ray-Ban smart glasses contains a complete, functional facial recognition system that's currently disabled for ordinary users. The discovery comes as Meta faces mounting criticism over privacy practices with its smart glasses technology.
Three ML Models and Local Vector Database Form Complete Pipeline
Buchodi's technical analysis uncovered a sophisticated architecture:
- Three ML models totaling approximately 100 MB: SCRFD for face detection, KPSAligner for face alignment, and SFace for generating 2048-dimensional biometric embeddings
- Local database (person_profiles/objects.db) with cosine-similarity vector indexing using sqlite-vec
- Storage mechanism (NameTagsPending/ directory) that stages unrecognized faces as cropped images plus fingerprints
- Notification pipeline configured with an Android channel called 'nametags_recognition'
When Buchodi directly triggered the system with test images, the pipeline successfully detected faces, generated embeddings, searched the local index, and fired notifications reading "Person recognized" with matched names. The user-facing "Connections" interface doesn't appear on stock accounts, and the profile screen the recognition notification links to is absent from this build version.
Architecture Suggests Intentional Engineering, Not Experimental Code
Buchodi noted that the embedding dimensions matching index specifications, complete write paths, and production-quality notification infrastructure suggests "intentional engineering rather than experimental code." The feature, internally called "Name Tag," would allow wearers to identify individuals and receive real-time information about them through Meta's AI.
Engineers considered two versions: recognition only for individuals already connected to the user on Meta platforms, or identification of anyone with a public account on social media services like Instagram.
Privacy Advocates Called for Halt in April 2026
The timeline of events:
- February 2026: New York Times reported Meta was considering adding facial recognition, targeting year-end launch
- April 2026: 70+ advocacy organizations called on Meta to halt the plans
- June 4, 2026: Buchodi's analysis published, generating 219 points and 196 comments on Hacker News
Meta spokesperson Ryan Daniels stated: "we've said before we're exploring these types of features, and what you're seeing is just evidence of that exploration" and "Nothing has shipped to consumers and no final decision has been made on what to do here, if anything."
Legal Context: Class-Action Lawsuit Already Pending
Meta's Ray-Ban smart glasses already face a federal class-action lawsuit alleging the company sold the product on explicit privacy promises while routing footage to human reviewers overseas through a pipeline buyers could not opt out of. Critics warn that facial recognition on smart glasses creates unprecedented surveillance capabilities, allowing wearers to identify strangers in public without consent. The technology could be used for stalking, harassment, or mass surveillance.
Key Takeaways
- Security researcher Buchodi discovered a complete, functional facial recognition system in Meta's Stella app for Ray-Ban smart glasses on June 4, 2026
- The system includes three ML models (SCRFD, KPSAligner, SFace), a local vector database with 2048-dimensional embeddings, and production-quality notification infrastructure
- When tested, the system successfully detected faces, generated embeddings, searched the local index, and fired "Person recognized" notifications
- Meta's Ray-Ban smart glasses already face a federal class-action lawsuit over privacy violations involving footage routing to human reviewers
- 70+ advocacy organizations called on Meta to halt facial recognition plans in April 2026, two months before the functional system was discovered