Skip to content

Security: mayflower/wortlaut

SECURITY.md

Security Policy

wortlaut joins confidential meetings, processes real-time audio streams, and handles OAuth2 / OIDC credentials. We treat security reports with the highest priority.

Supported Versions

Version Supported
main / latest 0.1.x release Yes

Reporting a Vulnerability

Please do NOT open a public GitHub issue for suspected security vulnerabilities.

Instead, please report vulnerabilities privately via GitHub Private Vulnerability Reporting:

Please include:

  1. Affected component (services/meeting, services/asr, worker, charts/wortlaut, or docker-compose.yml).
  2. Reproduction steps or proof-of-concept request/payload.
  3. Impact assessment (e.g., cross-tenant data access, token leakage, SSRF, container escape).

We aim to acknowledge reports within 2 business days and coordinate a fix and advisory before public disclosure.

Security Architecture Highlights

  • Ephemeral Worker Isolation: Every recording attempt runs in a dedicated, short-lived Kubernetes Job or unprivileged Docker container (CapDrop: ALL, no-new-privileges, scoped single-attempt bootstrap credential).
  • Restricted Docker Control Plane: In Docker Compose deployments, the meeting service runs as non-root (10001) without /var/run/docker.sock mounted; container lifecycle calls are mediated through docker-socket-proxy with PRIVILEGED=0, VOLUMES=0, and EXEC=0.
  • Scoped Read-Only Protocol Tokens: Shared protocol links (wortlaut-read-…) are HMAC-SHA256 bound to a single recording_id + per-recording rotatable nonce and grant read-only access to that recording's transcript exports only.
  • No Biometrics: Speaker attribution relies strictly on conference signaling and isolated WebRTC receiver tracks — never voiceprints or biometric speaker embeddings.

There aren't any published security advisories