Hidden facial recognition code in Meta AI

2026-06-05

Code, not policy statements, is now defining Meta’s privacy ambitions. Hidden fragments inside Meta’s AI assistant app, examined by Wired, reference an unreleased facial recognition feature that appears wired to Meta account data and device cameras, even though no such consumer tool has been launched.

What looks like a dormant option is in fact a design choice with clear intent. Strings discovered in the app describe enrolling faces, matching them, and linking results to user identifiers, suggesting a full verification and inference pipeline rather than a lightweight on-device filter. Wired’s inspection follows earlier reporting that Meta had explored facial recognition for its camera-equipped smart glasses, which already stream visual data to Meta’s servers for object detection and scene understanding.

The more striking point is how little public disclosure has kept pace with technical integration. Privacy rules, including data minimization and purpose limitation, hinge on whether biometric templates and feature vectors are generated, stored, or shared across products, yet Meta has not detailed any such architecture for this unreleased feature. The buried references highlight a familiar pattern in consumer AI: capabilities arrive first in code repositories and app binaries, while transparency, consent flows, and independent oversight trail behind by several product cycles.

Recommendations