AI Controls Management (e.g., Evaluation, Selection, Validation) AAIR Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A logistics company has implemented a control that automatically reroutes deliveries when the AI routing model predicts a high probability of delay. Before full deployment, the risk team wants to validate the control's impact on delivery times. Which validation method would provide the most reliable evidence?

   • AAsk delivery drivers for their opinion on the control's effectiveness
   • BDeploy the control in a single region and compare delivery times to other regions
   • CRun a simulation using historical delivery data to compare outcomes with and without the control
   • DReview the control's algorithm to ensure it is logically sound
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   At which stage of the AI lifecycle should controls be integrated to ensure they are most effective?

   • ADesign and development
   • BDeployment
   • CMonitoring
   • DRetirement
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A bank's AI risk team has selected a new control to monitor model drift in its loan approval system. The control will trigger alerts when performance degrades beyond a threshold. Which documentation and communication step is most important to ensure transparency and accountability?

   • APublish the control's source code on the internal developer wiki for transparency
   • BStore the control documentation in a shared drive without any formal notification
   • CDocument the control's purpose, threshold, and escalation path, and communicate this to model owners, compliance, and internal audit
   • DSend a one-time email to the model development team describing the control
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A manufacturing company is integrating AI controls into its predictive maintenance system. The system is being developed in phases: data collection, model training, deployment, and monitoring. Which approach best integrates controls across the entire AI lifecycle?

   • AEmbed control checkpoints at each phase: data quality checks during collection, fairness tests during training, rollback triggers at deployment, and drift detection in monitoring
   • BConduct a single comprehensive control review just before the system goes live
   • CApply all controls only during the monitoring phase to simplify management
   • DRely on the model development team to self-assess controls at each phase
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A multinational bank is selecting controls for its AI-based anti-money-laundering (AML) transaction monitoring system. The system must comply with regulations in multiple jurisdictions, each with different reporting requirements. The bank has a moderate risk appetite but faces significant operational costs from false positives. Which control selection approach best addresses these conflicting constraints?

   • ASelect a core set of controls that meet the most stringent regulatory requirements, then add jurisdiction-specific reporting overlays
   • BChoose a single global control that satisfies the least stringent jurisdiction to reduce costs
   • CImplement separate controls for each jurisdiction to ensure full compliance
   • DDefer control selection until all jurisdictions harmonize their regulations
   Reveal Solution Discussion Select an answer first