Bypassing Defensive Controls GOAA Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   An organization's AI-based network intrusion detection system (NIDS) has been evaded by adversarial traffic. The security team is considering two mitigations: adversarial training and input sanitization. The team has limited computational resources and must choose one. Which consideration should guide the decision?

   • AAdversarial training is more robust against a wide range of adversarial perturbations, but requires significant computational resources for retraining.
   • BInput sanitization is easier to implement and provides a permanent solution to adversarial attacks.
   • CInput sanitization is always more effective than adversarial training and requires no retraining.
   • DAdversarial training is the only mitigation that can prevent all evasion attempts.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   After a red-team exercise, an organization discovers that its AI-based intrusion detection system can be evaded by adversarial network traffic. The security team wants to propose a mitigation that is practical and effective. Which mitigation should they recommend?

   • ADisable the AI-based IDS until a new model is trained.
   • BIncrease the IDS's alert threshold to reduce false positives.
   • CImplement adversarial training by augmenting the training data with adversarial examples generated during the exercise.
   • DReplace the AI-based IDS with a traditional signature-based IDS.
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A security team has deployed an AI-based web application firewall that is vulnerable to adversarial SQL injection payloads. The team wants to mitigate the vulnerability without blocking legitimate queries that contain similar patterns. Which mitigation is most appropriate?

   • AUse a combination of input validation and adversarial training on the WAF model.
   • BBlock all queries that contain SQL keywords.
   • CIncrease the WAF's sensitivity to catch more malicious payloads.
   • DUse a simpler rule-based WAF instead of the AI model.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   An organization's AI-based email filter has been evaded by adversarial emails. The security team is considering implementing input sanitization, which removes suspicious characters and normalizes text. However, the team is concerned that sanitization may reduce the filter's ability to detect legitimate phishing emails. Which approach best balances security and usability?

   • ADisable sanitization to avoid any impact on detection, and rely on adversarial training instead.
   • BUse a separate model for sanitized and unsanitized emails, which doubles computational cost.
   • CApply aggressive sanitization that removes all non-alphanumeric characters, which may break legitimate emails.
   • DImplement sanitization that preserves the semantic content of emails while removing known adversarial patterns.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A security team is planning a red-team exercise to test an AI-based phishing detection system. The team wants to evaluate the system's robustness against adversarial emails. Which step should be performed first?

   • ADeploy the adversarial examples directly to the production email system.
   • BDisable the phishing detection system to avoid interference during testing.
   • CDefine the scope and success criteria for the exercise, including what constitutes a successful bypass.
   • DImmediately generate adversarial examples using a surrogate model.
   Reveal Solution Discussion Select an answer first