FaceLock is built with privacy-by-design principles:

1. 100% On-Device Processing:

   • All facial embeddings are computed locally on the Apple Neural Engine via Apple's Vision framework (VNCreateFaceprintRequest).
   • Camera video frames are processed in memory and never written to disk during regular verification.
   • Only when an unauthorized intruder triggers a verification failure is an encrypted JPEG snapshot recorded to ~/.facelock/intruders/ for local security audit logs.

2. No Cloud Telemetry or Tracking:

   • FaceLock contains zero telemetry, zero analytics SDKs, and zero background network calls.
   • Outbound network requests occur exclusively if you explicitly configure Telegram Bot alerts to send intruder notifications to your personal Telegram Bot token.

3. Biometric Representations:

   • FaceLock stores mathematical 128-dimensional float vectors (~/.facelock/owner_face_vectors.json). Raw face photos are discarded immediately after calibration.

If you discover a security vulnerability within FaceLock, please submit an advisory via GitHub Security Advisories or contact the repository maintainers directly. Please allow reasonable time for remediation before any public disclosure.