Incident Notification Procedure

Tenant

Acme Health Systems (t-acme-health)

Contact

security@acme-health.example

Audit window

2026-04-15 03:10:49 UTC → 2026-05-15 03:10:49 UTC

Generated

2026-05-15 03:10:49 UTC

What constitutes a notifiable incident

• Data plane. Confirmed unauthorized access to the gateway's runtime memory or audit store. Confirmed compromise of API key validation. Confirmed exfiltration of audit metadata.
• Control plane. Confirmed unauthorized administrative access to Optimal's deploy infrastructure. Confirmed CMK material compromise. Confirmed CloudTrail tampering or gap.
• Upstream provider. Confirmed incident at Bedrock / Vertex / AOAI that affects the customer's traffic. Optimal forwards the upstream notification within 24h.

What is NOT a notifiable incident

• Customer-side compromise (your IdP, your app server, your API key custody).
• Customer's content quality issues (model output disputes, prompt-injection attempts that the gateway successfully blocked).
• Routine maintenance events covered by the change-management procedure.

Notification SLA

Optimal will notify the customer contact on file (security@acme-health.example) within 24 hours of incident confirmation. Tier upgrades shorten this SLA.

Notification content

• Incident classification (data plane / control plane / upstream).
• Time of detection and time of confirmation.
• Affected scope (audit_id range, api_key_id list, model/provider).
• Forensic metadata available for customer review. Note: prompt + response bodies do not exist at rest within the gateway boundary; "forensic data" in this context is metadata.
• Containment status and next steps.

Customer responsibilities during incident

• Rotate the affected API key(s) on receipt of the notification.
• Quiesce traffic through the gateway if your incident response plan calls for it.
• Provide a customer-side contact who can respond within your own internal SLA.