AI Attack Taxonomies and Classification Models COASP Practice Questions (Page 7)

Questions 31–35

1. 31foundation · easy

   A vulnerability that allows an attacker to query a model and reconstruct its decision boundary is best categorized under which AI component?

   • AInference
   • BModel
   • CData
   • DInfrastructure
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A security team is creating a threat model for an AI system and wants to use a taxonomy to ensure they cover all major attack categories. They have identified attacks that occur during the training phase and attacks that occur during the inference phase. Which of the following correctly maps the attack types to the phases?

   • APoisoning attacks occur during training; evasion attacks occur during inference
   • BEvasion attacks occur during training; poisoning attacks occur during inference
   • CExtraction attacks occur during training; inference attacks occur during inference
   • DAll attacks occur during both training and inference
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A security analyst is investigating a breach of a facial-recognition system. The analyst finds that the attacker submitted many photos of themselves to a public API and used the responses to build a copy of the model's decision function. Which attack category and AI component does this represent?

   • AExtraction attack targeting the model component
   • BEvasion attack targeting the inference component
   • CInference attack targeting the data component
   • DPoisoning attack targeting the data component
   Reveal Solution Discussion Select an answer first
5. 34application · medium

   A security analyst at a financial firm discovers that an attacker has been submitting carefully crafted loan applications to a credit-scoring model. Each application is designed to push borderline applicants just over the approval threshold. The attacker's goal is to get more loans approved, not to disrupt the service. Which attack category and AI component does this scenario best represent?

   • APoisoning attack targeting the training data pipeline
   • BInference attack targeting the training data's privacy
   • CEvasion attack targeting the model's decision boundary at inference time
   • DExtraction attack targeting the model's internal parameters
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A machine learning engineer is reviewing the security of a spam filter. The team discovers that an attacker has been submitting specially crafted emails that, when used as training data, cause the filter to misclassify legitimate emails as spam. The attacker's goal is to degrade the overall accuracy of the model. Which vulnerability category and attack type does this scenario represent?

   • AModel vulnerability; evasion attack
   • BData vulnerability; inference attack
   • CData vulnerability; poisoning attack
   • DInference vulnerability; extraction attack
   Reveal Solution Discussion Select an answer first