Skip to content

Security And Reliability Incident Response

Severity Examples Initial response target
SEV-0 Active credential theft, authorization bypass, unredacted sensitive-data exposure, audit integrity loss 15 minutes
SEV-1 Governed writes executing incorrectly, widespread outage, unrecoverable audit/cache barrier failure 30 minutes
SEV-2 Partial source/connector outage, worker backlog, degraded discovery, bounded audit spool growth 4 hours
SEV-3 Non-security defect with workaround, documentation or certification gap 1 business day

Targets are operator objectives and require an on-call process.

  1. Assign incident commander, operations lead, security lead, and recorder.
  2. Capture time, release SHA, image digest, affected components, correlation IDs, and safe symptoms.
  3. Contain exposure:
    • disable affected source or identity,
    • revoke/rotate compromised credentials,
    • block connector egress,
    • pause approvals and ingestion,
    • reject governed traffic if audit integrity is uncertain.
  4. Preserve logs, traces, audit partitions, dead letters, spool volumes, manifests, and database snapshots read-only.
  5. Do not paste credentials, raw PII, or approval payloads into tickets/chat.
  • Did source roles, policy, rate limit, write safety, redaction, cache, and audit produce the same decision for the affected request?
  • Was the identity or grant recently changed or cached?
  • Did a cache key cross source, identity, role, grant, or policy scope?
  • Did egress validation or DNS resolution change between validation/connect?
  • Did audit enter the database, spool, retry queue, or dead-letter path?
  • Did a Worker lose its lease or complete after cancellation?
  • Is the deployed digest the certified digest?
  • Were migrations and configuration identical across replicas?
  • Authorization/redaction: disable source/identity, stop Gateway traffic, invalidate source cache generation, preserve audit, patch before resume.
  • Audit unavailable: strict mode rejects new governed work; preserve spools, restore PostgreSQL, replay idempotently, verify partition health.
  • Redis/cache: reject writes when the consistency barrier is unavailable; restore Redis, advance source generations, invalidate affected source caches.
  • Worker/ingestion: pause claims, expire/reclaim leases deliberately, quarantine poison jobs, verify source asset versions before replay.
  • Credential exposure: revoke at provider first, rotate InterLock references, search audit/log/UI artifacts, notify affected owners.

Recovery requires explicit incident-command approval and:

  • Root cause or safe containment is understood.
  • Compromised credentials are rotated.
  • Fixed artifact is reviewed and identified by digest.
  • Audit continuity is reconciled.
  • Cache generations are safe.
  • Smoke and adversarial regression tests pass.
  • Monitoring is heightened for a defined observation period.

Use the process in the repository SECURITY.md for vulnerability disclosure. Public communication must state affected versions, impact, mitigations, fixed version, and upgrade instructions without exposing customer or source details.

Within five business days for SEV-0/1, publish an internal review covering timeline, detection gap, control behavior, contributing factors, corrective owners/dates, regression tests, and runbook changes. Security advisories should be public when users need action.