AI Assurance, Testing, and Auditing CRAGE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A retail company has deployed a recommendation engine. After a major product catalog update, the model's performance metrics (e.g., click-through rate) have declined. The data science team suspects the model is now receiving inputs outside its training distribution. Which ongoing assurance practice would have most likely detected this issue early?

   • AMonitor the model's input feature distributions and compare them to the training distribution in real time.
   • BReview the model's explainability reports generated at deployment time.
   • CPerform a one-time post-deployment audit of the model's training data for bias.
   • DRe-run the model's robustness tests on the original test set after the catalog update.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   What is the primary role of AI assurance in an organization's AI governance program?

   • ATo ensure that AI models achieve the highest possible accuracy on every dataset.
   • BTo guarantee that AI systems are free from all biases and errors.
   • CTo replace all manual testing with automated tools that run continuously without human oversight.
   • DTo provide stakeholders with confidence that AI systems operate within acceptable risk and compliance boundaries.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   What does 'interpretability' in AI refer to?

   • AThe degree to which a human can understand the cause of a model's decision
   • BThe ability of the model to learn from new data
   • CThe model's accuracy on unseen data
   • DThe speed at which the model makes predictions
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A financial firm is auditing an AI system that approves loan applications. The audit team must prioritize risks based on likelihood and impact. They have limited time and must focus on the most critical risk. Which approach best aligns with a standard risk-assessment framework?

   • AFocus only on risks with the highest likelihood, ignoring impact.
   • BRank risks by the product of likelihood and impact, then address the highest-scoring risk first.
   • CDefer all risk assessment until after the audit report is written.
   • DAddress all identified risks equally regardless of likelihood or impact.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   Which standard is specifically designed to provide requirements for establishing, implementing, maintaining, and continually improving an AI management system?

   • AISO/IEC 20000
   • BISO/IEC 27001
   • CISO/IEC 42001
   • DISO/IEC 27701
   Reveal Solution Discussion Select an answer first