# AI support incident runbook
Condition and severity:
Owner and backup:
Affected environment, conversations and operation IDs:
Pause new public replies and writes when required by the incident policy.
Preserve receipts, logs and reconciliation work.
Identify the failing dependency and the actual running configuration.
Reconcile unknown business actions before retrying.
Use only an evaluated fallback; otherwise use the approved customer message and human route.
Restore service with a narrow check of downstream state.
Record impact, recovery evidence, root cause and regression cases.
Review data retention and access if sensitive information was involved.