Artificial Intelligence Fundamentals GASAE Practice Questions (Page 6)

Questions 26–30

1. 26expert · hard

   A red team is tasked with evaluating the security of an AI-based authentication system that uses behavioral biometrics. The team has access to the system's API but not its internal model. They want to test whether an attacker can impersonate a legitimate user. Which approach best tests this?

   • APoison the training data by submitting fake behavioral samples
   • BCraft adversarial behavioral samples using a substitute model and test them against the API
   • CDirectly access the model's weights to craft a white-box attack
   • DPerform a denial-of-service attack on the API
   Reveal Solution Discussion Select an answer first
2. 27application · medium

   A red team is conducting an adversary emulation exercise against a company's AI-based security automation platform. The goal is to test whether the platform's automated response actions can be tricked into performing harmful actions. Which technique best simulates a realistic adversary?

   • ASend a high volume of normal traffic to see if the platform becomes overwhelmed.
   • BCraft inputs that cause the AI to misclassify a benign action as malicious, triggering a disruptive response.
   • CUse a known malware sample to see if the platform detects it.
   • DDisable the AI model and test the platform's manual response procedures.
   Reveal Solution Discussion Select an answer first
4. 28expert · hard

   A security team is evaluating two AI-based intrusion detection systems. System A has a high detection rate but generates many false positives. System B has a lower detection rate but very few false positives. The team's SOC has limited staff and cannot afford to investigate many false alarms. Which system should they choose, and why?

   • ASystem A, because false positives are harmless and can be ignored
   • BSystem A, because detection rate is always the most important metric
   • CSystem B, because it will catch more attacks overall
   • DSystem B, because the cost of false positives is higher than the cost of missed detections in this context
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   A SOC wants to reduce the time to detect and respond to incidents. They plan to use AI to automate the correlation of alerts from multiple sources and suggest response actions. Which AI capability is most directly relevant?

   • AUser interface design for the SOC dashboard
   • BPattern recognition and automated decision support
   • CData compression to reduce storage costs
   • DNetwork traffic encryption
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A security team has deployed an AI model that classifies URLs as malicious or benign. They want to measure how well the model performs on real-world data, where malicious URLs are rare. Which evaluation metric is most appropriate to assess the model's ability to correctly identify malicious URLs without overwhelming analysts with false alarms?

   • AAccuracy, because it is the most straightforward metric.
   • BArea under the ROC curve (AUC) only, because it is threshold-independent.
   • CF1 score alone, without considering the operational cost of false positives.
   • DPrecision and recall, with a focus on recall for malicious URLs.
   Reveal Solution Discussion Select an answer first