Adversarial Attacks CT-AI Practice Questions (Page 5)

Questions 21–24

1. 21foundation · easy

   What is a potential impact of an adversarial attack on an AI-based system's reliability?

   • AThe system may produce incorrect outputs on inputs that are slightly modified, leading to unpredictable behavior.
   • BThe system will always crash when it encounters an adversarial example.
   • CThe system will become slower because it needs to process the adversarial input.
   • DThe system's training data will be permanently corrupted.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A company deploys an AI-based spam filter. They want to defend against evasion attacks that add subtle perturbations to emails to bypass the filter. Which defense mechanism is most directly applicable?

   • ARate limiting API requests
   • BAdversarial training
   • CModel ensembling
   • DInput preprocessing (e.g., sanitization)
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   What is the primary characteristic of an adversarial attack on an AI-based system?

   • AAn attack that exploits a vulnerability in the underlying network infrastructure to gain unauthorized access to the AI model.
   • BAn attack that involves feeding the model with inputs that are intentionally crafted to cause it to make a mistake, while appearing normal to humans.
   • CAn attack that overwhelms the AI system with a high volume of requests to cause a denial of service.
   • DAn attack that steals the training data used to build the AI model by exploiting a SQL injection vulnerability.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A security team is assessing the threat model for an AI-based image recognition system used for physical access control. The system is deployed in a controlled environment with no internet access, and the training data is proprietary. The team wants to prioritize defenses against the most realistic threat. Which threat model should they consider as the highest priority?

   • AAn attacker with white-box access to the model who can craft physical adversarial patches
   • BAn attacker who can extract the model through API queries
   • CAn attacker who can poison the training data through public data sources
   • DAn external attacker with black-box access who can query the model remotely
   Reveal Solution Discussion Select an answer first