Incident Response Management Specific to AI AAIA Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A manufacturing company's predictive maintenance model was found to be trained on corrupted sensor data, causing it to miss critical equipment failures. The incident response team has contained the issue by switching to manual monitoring. What is the NEXT step in the eradication and recovery phase?

   • ARetrain the model using corrected and validated sensor data.
   • BDocument the incident and move on without further action.
   • CPermanently disable the model and rely on manual monitoring.
   • DIncrease the frequency of manual inspections indefinitely.
   Reveal Solution Discussion Select an answer first
2. 32foundation · easy

   Which of the following is a containment strategy for an AI incident?

   • AIgnoring the incident until the next scheduled maintenance
   • BRolling back the model to a previous version
   • CIncreasing the model's learning rate
   • DDeleting all training data
   Reveal Solution Discussion Select an answer first
4. 33expert · hard

   After a major AI incident, the post-mortem reveals that the incident was not detected earlier because the monitoring system only tracked overall model accuracy, not per-segment performance. The team wants to improve detection. Which improvement is MOST aligned with the lessons learned?

   • AImplement per-segment monitoring to detect degradation in specific user groups or data subsets.
   • BDisable the model and rely on rule-based systems.
   • CAdd more human reviewers to manually check model outputs.
   • DIncrease the frequency of model retraining to prevent degradation.
   Reveal Solution Discussion Select an answer first
5. 34foundation · easy

   Which of the following is an example of input filtering as a containment measure?

   • AIncreasing the model's batch size
   • BBlocking inputs that match known adversarial patterns
   • CChanging the model's activation function
   • DRestarting the model server
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A healthcare AI system that predicts patient readmission risk has been producing increasingly inaccurate predictions over the past week. The data science team suspects data drift but has not confirmed it. The incident response team needs to detect and confirm the issue. Which approach is MOST appropriate for early detection?

   • ACheck the model's accuracy on a static test set from the original training period.
   • BReview the model's training code for bugs.
   • CMonitor the model's prediction distribution against the training data distribution using statistical tests.
   • DWait for user complaints to confirm the issue.
   Reveal Solution Discussion Select an answer first