# Agent incident communications plan

Use this plan when an AI service creates a user-visible incident: wrong action, unsafe answer, data exposure, degraded service, policy violation, or support escalation failure. The plan keeps communication disciplined while engineering and operations teams investigate.

## Communication matrix

- Internal first line: incident commander, engineering owner, support owner, risk owner, and executive sponsor.
- Customer update: affected audience, known impact, current containment, next update time, and support contact.
- Regulatory or contractual notice: trigger condition, responsible approver, deadline, evidence needed, and record location.
- Postmortem: timeline, root cause, customer impact, detection gap, remediation, owner, and completion date.

## Message standards

Avoid speculation. State what is known, what is being verified, what has been contained, and when the next update will arrive. Every communication should map to an incident record and a responsible owner.