AI incident response

AI incident response follows a fixed flow: detection (guardrail event, complaint or anomaly) → severity classification (S4 low to S1 critical) → containment (kill switch, demotion or token revocation) → reverse or compensate affected Actions per their reversibility class → notify the tenant, and authorities if a personal-data breach requires it → root cause by replaying the run → fix plus new eval cases → post-mortem → Knowledge.

Severity sets the automatic response: an S3 demotes the skill one level; an S2 suspends the skill and external channels; an S1 triggers a kill switch and freezes related Actions. The Agent Liability Desk for claims and remediation is planned for W4.