Mitigate Adversarial Risks CY0-001 Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A security vendor sells an AI-based intrusion detection system. A customer reports that the system is failing to detect certain attacks that were previously detected. The vendor suspects that the customer's network traffic has been subtly modified to evade the model. Which defense would be most effective in this scenario, considering the need to maintain detection performance?

   • AUse a signature-based detection method instead of the AI model.
   • BUse adversarial training with a diverse set of evasion techniques to improve the model's robustness.
   • CDeploy the model on more powerful hardware to process traffic faster.
   • DIncrease the model's sensitivity by lowering the detection threshold.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A bank uses an AI model to detect fraudulent transactions. The security team wants to implement a defense that does not require retraining the model and can be applied at inference time to filter out adversarial inputs. Which approach fits this requirement?

   • AApply input preprocessing techniques such as feature squeezing or smoothing.
   • BRetrain the model with adversarial examples generated from the current model.
   • CIncrease the model's decision threshold to reduce false positives.
   • DUse a larger training dataset to improve the model's generalization.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A defense contractor is developing an AI system for threat detection. The project has strict security requirements, and the team must ensure that the system is resilient to adversarial attacks. During threat modeling, they identify that the training data could be poisoned by a malicious insider. Which combination of controls would best mitigate this risk?

   • ADeploy the model behind a firewall to protect it from external attacks.
   • BUse a more complex model to better distinguish between legitimate and poisoned data.
   • CUse a larger training dataset to reduce the impact of any poisoned data.
   • DImplement strict access controls on the training data and use data provenance tracking.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   How can explainability techniques help in detecting adversarial influences on AI models?

   • ABy automatically retraining the model to be more robust
   • BBy encrypting the model's outputs
   • CBy increasing the model's accuracy on clean data
   • DBy revealing which input features most influenced a model's decision, making it easier to spot unusual patterns
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A large online platform uses an AI model to detect fake accounts. The security team has implemented monitoring that tracks the model's prediction confidence. They notice that the average confidence for new accounts has dropped significantly over the past week, while the false positive rate has increased. They suspect an adversarial campaign. Which action would best help them confirm the attack and understand its nature?

   • ABlock all new account registrations temporarily to stop the attack.
   • BIncrease the model's decision threshold to reduce false positives.
   • CAnalyze the feature attributions for the flagged accounts using SHAP to identify common patterns.
   • DImmediately retrain the model with the latest data to improve its accuracy.
   Reveal Solution Discussion Select an answer first