AI Vulnerability Scanning and Fuzzing COASP Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which of the following is an example of generating an adversarial input sample for an AI model?

   • AEncrypting the model's weights
   • BChanging the model's learning rate
   • CAdding imperceptible noise to an image to cause misclassification
   • DRebooting the inference server
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   Which type of tool is specifically designed for fuzzing AI models?

   • AA tool that generates adversarial examples for deep learning models
   • BA vulnerability scanner like Nessus
   • CA web application scanner like OWASP ZAP
   • DA general-purpose network fuzzer like AFL
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   What does a high rate of misclassifications in a fuzzing campaign indicate?

   • AThe model is performing optimally
   • BThe training data was too small
   • CThe model is likely vulnerable to adversarial inputs
   • DThe fuzzer is not working correctly
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A team is fuzzing a speech-to-text model to find inputs that cause it to transcribe harmful content or fail catastrophically. They have a corpus of normal audio clips. Which input generation strategy would be most effective for exploring the model's edge cases?

   • AOnly use the original corpus without modification to test the model's baseline accuracy.
   • BConvert the audio to text and fuzz the text before sending it to the model.
   • CFeed the model random noise audio clips of varying lengths.
   • DApply small perturbations to the audio waveforms, such as adding imperceptible background noise or slight speed changes, and also generate adversarial examples using gradient-based methods.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A team is fuzzing a sentiment analysis model. They have a set of seed sentences and want to generate inputs that are likely to cause misclassification. Which input generation approach is most aligned with AI fuzzing best practices?

   • AUse a fuzzer that only performs random byte-level mutations on the input tokens.
   • BUse a grammar-based generator to produce random sentences from a context-free grammar.
   • CApply word-level mutations such as replacing words with synonyms, adding typos, and inserting adversarial phrases, guided by the model's gradients.
   • DGenerate inputs by randomly shuffling the characters in the seed sentences.
   Reveal Solution Discussion Select an answer first