AI Threat and Vulnerability Management AAISM Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   Which step in the AI vulnerability management lifecycle involves confirming that a remediation has been successfully applied?

   • ARemediation
   • BIdentification
   • CPrioritization
   • DVerification
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A company's image recognition model is vulnerable to adversarial examples that cause misclassification. The security team wants to implement a mitigation that is effective and has minimal impact on model performance. Which mitigation is most appropriate?

   • AImplement adversarial training by augmenting the training data with adversarial examples.
   • BUse a different model architecture that is inherently robust to adversarial examples.
   • CAdd a preprocessing step that applies a median filter to all input images.
   • DRestrict access to the model so only authenticated users can make predictions.
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A government agency is developing a threat model for a new AI system that classifies citizen requests. The system uses a third-party pre-trained language model and is accessed by the public via a web portal. Which element is most important to include in the threat model to address supply chain risk?

   • AThe user authentication mechanism for the web portal.
   • BThe model's accuracy on different demographic groups.
   • CThe physical security of the agency's data center.
   • DThe trust boundary between the third-party model provider and the agency's infrastructure.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A security team is designing monitoring for an AI system that classifies loan applications. The team wants to detect if an attacker is attempting to infer sensitive information about applicants through model inversion. Which monitoring approach is most effective?

   • AMonitor the distribution of model outputs and alert on unusual patterns that suggest repeated queries for the same or similar inputs.
   • BMonitor the system's CPU usage to detect resource-intensive attacks.
   • CMonitor the model's training data for signs of poisoning.
   • DReview the model's confidence scores for anomalies.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is the purpose of monitoring AI system behavior for anomalies?

   • ATo detect potential attacks or malfunctions in real time
   • BTo reduce the model's inference latency
   • CTo automatically retrain the model on new data
   • DTo improve the model's accuracy on standard benchmarks
   Reveal Solution Discussion Select an answer first