AI Security Testing, Evaluation, and Hardening COASP Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   Which of the following is a widely recognized framework for AI security best practices?

   • AOWASP Top 10 for Large Language Model Applications
   • BPCI DSS
   • CISO 27001
   • DNIST SP 800-53
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Which framework provides a collection of attack and defense algorithms for evaluating AI model robustness?

   • AAdversarial Robustness Toolbox (ART)
   • BTensorFlow Privacy
   • CScikit-learn
   • DOpenAI Gym
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   Which methodology is commonly used in AI security testing to evaluate a model's behavior under malicious input perturbations?

   • AAdversarial testing
   • BUnit testing
   • CFuzz testing
   • DLoad testing
   Reveal Solution Discussion Select an answer first
5. 29expert · hard

   A government agency uses an AI system to analyze satellite imagery for border security. The system has been compromised by an adversary who has inserted a backdoor that triggers when specific patterns are present in the imagery. The agency needs to respond to this incident while maintaining its operational capabilities. Which action should the agency take first?

   • ANotify the public about the compromise to ensure transparency, as required by government regulations.
   • BIsolate the compromised AI system from the operational network and switch to manual analysis of satellite imagery until the backdoor is removed.
   • CImmediately retrain the model on a new dataset that does not contain the backdoor trigger patterns.
   • DDeploy the model with additional monitoring to detect any future backdoor activations.
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A security engineer is testing a fraud-detection model that uses tabular transaction data. The team wants to know if an attacker who has access to the model's confidence scores (but not the training data) could reconstruct the training set. Which attack simulation should the engineer run first?

   • AModel inversion attack using the confidence scores to infer feature values of training records
   • BMembership inference attack to determine if a specific record was in the training set
   • CGradient-based evasion attack to craft adversarial examples that bypass the model
   • DData poisoning attack by injecting malicious samples into the training pipeline
   Reveal Solution Discussion Select an answer first