Knowledge MaxxingIncident reporting
Incident reporting is a defined process for recording and escalating when an AI system causes unexpected harm or behaves outside its intended bounds.
Prerequisite: AI audit
What matters
Without a clear reporting process, harmful incidents can go unnoticed or unaddressed, since no one is specifically responsible for surfacing and acting on them.
Example
A defined incident-reporting process might require any employee who notices a model making a clearly biased decision to file a report that triggers a formal review.
Common misconception
Incident reporting is not just for catastrophic failures; smaller unexpected behaviors are often worth reporting too, since patterns across minor incidents can reveal a bigger issue.
Recall question
Why is it valuable to report smaller unexpected incidents, not just major failures?