AI Vulnerability Scanning and Fuzzing COASP Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   What is the primary purpose of vulnerability scanning in AI systems compared to traditional software scanning?

   • ATo detect weaknesses in the AI model's behavior, data, and inference logic that could be exploited by adversarial inputs
   • BTo ensure the AI model is deployed on the latest version of the cloud provider's hardware
   • CTo verify that the AI model meets its accuracy benchmarks on the training dataset
   • DTo identify misconfigurations in the underlying operating system and network services
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A security team is planning to fuzz a natural language processing (NLP) model that classifies customer support tickets. They want to uncover edge cases where the model produces incorrect or unsafe outputs. Which approach best reflects AI-specific fuzzing as opposed to traditional fuzzing?

   • ARun a coverage-guided fuzzer on the model's underlying C++ library to find memory corruption bugs.
   • BMutate valid ticket texts by inserting typos, synonyms, and adversarial phrases, and monitor changes in classification confidence and output labels.
   • CUse a standard HTTP fuzzer to send malformed JSON payloads to the API endpoint.
   • DGenerate random byte sequences and send them to the model's API to check for crashes.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A scan of a medical diagnosis AI system reveals that the model is vulnerable to model inversion attacks, allowing an attacker to reconstruct patient images from the model's outputs. What is the most effective mitigation?

   • ARetrain the model with a larger dataset to improve generalization.
   • BLimit the number of API queries per user and add noise to the model's output probabilities.
   • CDisable the API and only allow batch processing of medical images.
   • DEncrypt the model's output responses in transit.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A team is fuzzing a large language model (LLM) that is used for code generation. They want to find prompts that cause the model to generate insecure code. They have a set of seed prompts and want to use a specialized fuzzing framework. Which approach is most effective?

   • AOnly test the seed prompts without any mutation.
   • BUse a framework like LangFuzz that mutates prompts based on language grammar and evaluates the generated code for security issues.
   • CRandomly concatenate words from a dictionary to create prompts.
   • DUse a general-purpose fuzzer like AFL on the model's input API.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A security analyst is reviewing the results of a scan on a machine learning model. The scan indicates that an attacker could potentially reconstruct approximate versions of training data by querying the model. Which vulnerability category does this finding represent?

   • AAdversarial perturbation
   • BData poisoning
   • CMembership inference
   • DModel inversion
   Reveal Solution Discussion Select an answer first