Meta's Ray-Ban smart glasses app contains fully developed code for facial recognition, capable of identifying known individuals. This feature, currently inactive, uses a 2,048-dimensional biometric fingerprinting system, raising immediate privacy flags. The capability is built and wired down to a "Person recognized" notification, suggesting deployment is a simple switch away.
How We Got Here
Meta's history of handling sensitive user data from these glasses is already concerning. In March 2026, an investigation revealed contractors accessed unanonymized user data, including nudity, from live streams.
The Numbers
- The facial recognition pipeline consists of three models totaling around 100 MB.
- These models are named `androidfacerecscrfd` (detection), `androidfacereckpsaligner` (cropping), and `androidfacerec_sface` (embedding).
- The system converts a detected face into a unique 2,048-dimensional embedding, acting as a biometric fingerprint.
- Buchodi, a security-minded engineer, conducted the code inspection and published findings.
- The underlying technology leverages open-source architectures like InsightFace and SFace.
What Happens Next
🇮🇳 Why This Matters for India
For Bangalore's tech founders building consumer hardware, this highlights the tightrope walk between innovation and privacy ethics, especially with biometric data.
The Take
What's being missed is the sheer engineering effort: this isn't an accidental leftover; it's a deliberate design choice that Meta will eventually activate, perhaps with an opt-in. The company built and shipped this capability for a reason, and a future rollout, even if limited initially, looks inevitable.
Source:
MediaNama ↗