EthicaAISafety evaluationRisk register

AI incident response playbook

AI incidents need product, engineering, risk, legal, and support coordination. This playbook keeps the first response concrete.

Triage

Classify severity, affected users, active harm, model or data scope, and whether containment is required.

Contain

Pause a feature, disable a model route, add human review, roll back a version, or narrow access.

Learn

Record root cause, evidence, user impact, corrective controls, and monitoring updates.

Review governance controls