Threats and Vulnerabilities Specific to AI AAIA Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   An attacker repeatedly queries a machine learning model and uses the responses to build a functional copy of the model. This attack is known as:

   • APrompt injection
   • BModel extraction
   • CData poisoning
   • DModel inversion
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A bank deploys a deep learning model to approve credit card transactions. An attacker discovers that by adding a subtle, imperceptible pattern to a transaction image, the model consistently misclassifies fraudulent transactions as legitimate. The attacker exploits this flaw repeatedly without needing to alter the model's training data. Which type of AI-specific threat is being exploited?

   • AAdversarial attack
   • BModel inversion
   • CData poisoning
   • DPrompt injection
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   A backdoor attack on a machine learning model is a type of data poisoning. What is the defining characteristic of a backdoor attack?

   • AThe model's training data is completely deleted
   • BThe model's accuracy degrades uniformly across all inputs
   • CThe model performs well on normal inputs but misbehaves when a specific trigger is present
   • DThe model's architecture is altered to include malicious code
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A company deploys an LLM-based customer support system. They want to reduce the risk of prompt injection attacks that trick the model into revealing sensitive data. Which control would be most effective?

   • ARestricting model access to sensitive data
   • BInput validation and sanitization
   • COutput filtering
   • DModel retraining
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Why is lack of explainability considered a vulnerability in production AI systems?

   • AIt prevents the model from being deployed on cloud platforms
   • BIt increases the model's computational cost
   • CIt makes it difficult to detect and diagnose incorrect or biased model behavior
   • DIt makes the model more susceptible to adversarial attacks
   Reveal Solution Discussion Select an answer first