Artificial Intelligence Fundamentals GASAE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A company uses an AI model to classify files as benign or malicious. An attacker gains access to the training data pipeline and injects a small number of malicious files labeled as benign. After retraining, the model fails to detect these specific malicious files. Which type of attack occurred?

   • AModel inversion
   • BEvasion attack
   • CData poisoning
   • DPrompt injection
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A security analyst wants to use AI to detect anomalous user behavior that might indicate a compromised account. The analyst has a large dataset of normal user activity but very few examples of compromised accounts. Which machine learning approach is most suitable?

   • AReinforcement learning with a reward for detecting intrusions
   • BUnsupervised anomaly detection to identify deviations from normal behavior
   • CSupervised classification using the few compromised examples as the positive class
   • DA neural network trained only on compromised examples
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A security team is testing an AI-based malware detector. They want to evaluate its robustness against evasion attacks. They have access to the model's architecture and weights. Which evaluation method is most appropriate?

   • AEvaluate the model on a standard test set of benign and malicious files
   • BGenerate adversarial examples using white-box gradient-based methods and measure the model's accuracy on them
   • CMeasure the model's inference latency under load
   • DUse a black-box substitute model to generate adversarial examples
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A company uses an AI model to filter spam emails. An attacker wants to make the model misclassify a specific phishing email as benign. The attacker has no access to the model's training data or parameters, but can send test emails and observe whether they are flagged. Which attack type is the attacker most likely using?

   • ABlack-box evasion attack, where the attacker queries the model and crafts inputs based on the outputs.
   • BModel inversion attack, where the attacker reconstructs private training data.
   • CMembership inference attack, where the attacker determines if a sample was in training data.
   • DWhite-box poisoning attack, where the attacker modifies the training data.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A security automation team is evaluating whether to use a neural network or a traditional rule-based system for classifying malware samples. The team has a large labeled dataset of malware and benign files, and the malware evolves frequently. Which factor most strongly favors using a neural network?

   • AThe requirement to run on very limited hardware with minimal latency.
   • BThe need to classify samples without any historical data.
   • CThe ability to automatically learn complex patterns from raw features without manual rule updates.
   • DThe need for full transparency in every classification decision.
   Reveal Solution Discussion Select an answer first