AI Security Testing, Evaluation, and Hardening COASP Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A logistics company uses an AI model to optimize delivery routes. The model has been found to be vulnerable to a poisoning attack where an adversary has injected fake traffic data into the training set, causing the model to suggest inefficient routes. The company needs to recover the model's performance while minimizing the impact on ongoing operations. Which approach should the company take?

   • AApply adversarial training to the model to make it more robust to future poisoning attempts.
   • BSwitch to a completely new route-optimization algorithm that does not use AI, to avoid the risk of future attacks.
   • CDeploy the model immediately and monitor its route suggestions for any anomalies, assuming the attack will not recur.
   • DIdentify and remove the poisoned data samples from the training set, then retrain the model on the cleaned data while using a shadow model for ongoing operations.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which type of adversarial attack involves an attacker injecting malicious data into the training set to compromise the model's behavior?

   • AEvasion attack
   • BPoisoning attack
   • CExtraction attack
   • DInversion attack
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   What is the first step in an incident response plan tailored to AI systems?

   • ARecovery of the model to a known-good state
   • BEradication of the root cause
   • CContainment of the affected model
   • DDetection and identification of the AI-specific incident
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A financial institution must demonstrate to regulators that its credit-scoring model is robust against adversarial manipulation. The team has already run adversarial simulations and documented the results. What is the MOST important next step to satisfy compliance requirements?

   • AEstablish a formal testing policy that defines the scope, frequency, and methodology for adversarial testing
   • BIncrease the model's accuracy on the original test set
   • CShare the adversarial test results with all employees
   • DDeploy the model to production immediately since testing is complete
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A security team is evaluating a face-recognition model. They have access to the model's API but not its weights or training data. They want to determine if the model is vulnerable to extraction attacks. Which approach is most feasible given their access level?

   • AAnalyze the training data to find patterns that cause misclassification
   • BUse the API to query the model with crafted inputs and train a surrogate model to mimic its outputs
   • CInspect the model's weights directly to identify vulnerabilities
   • DRun a gradient-based attack that requires internal model gradients
   Reveal Solution Discussion Select an answer first