Question 840
At an enterprise maintaining a deployed AI service, an AI project leader faces this situation: A fraud classifier's predictions change after customer behavior shifts but nobody tracks feature distributions. Governance and technical measures need accountable ongoing operation. What result should the project owner require before accepting the change?
Explanation
Correct Answer
In an enterprise maintaining a deployed AI service, the decisive check is: A controlled distribution shift produces an alert tied to an investigation workflow. That shows the planned action has worked: Define reference distributions, drift metrics and alert thresholds for important inputs and outputs. The governing reason is: Input and performance drift can degrade model reliability after the initial acceptance evaluation. A separate successful activity would not establish this specific requirement.
Incorrect Answers
Inference behavior can change materially through configuration even without a new model artifact. This alternative mechanism does not match the current fault.
Operational governance should reassess obligations when scope, data sources or geography changes. Fixing that different mechanism would not address the current root cause: input and performance drift can degrade model reliability after the initial acceptance evaluation.
AI failures need predefined containment, triage, escalation and recovery actions suited to possible harm. Its verification is distinct from the reported condition: A fraud classifier's predictions change after customer behavior shifts but nobody tracks feature distributions. Governance and technical measures need accountable ongoing operation.