AI Reconnaissance and Attack Surface Mapping COASP Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   You are mapping the external attack surface of an AI service that offers both a web UI and an API. You have discovered the API base URL. Which next step would be most effective in discovering additional API endpoints?

   • ASend a GET request to the base URL and analyze the response body.
   • BInspect the web UI's JavaScript files for API calls and endpoints.
   • CRun a vulnerability scanner against the API base URL.
   • DPerform a DNS enumeration to find subdomains that may host additional APIs.
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   While mapping an AI service, you find a public API endpoint that returns model predictions. You also notice that the API accepts a 'model_version' parameter. What is the most important next step to fully map the attack surface?

   • AEnumerate all valid model versions and test each one for different behaviors or vulnerabilities.
   • BAttempt to access the model registry directly using the version number.
   • CIgnore the parameter and focus on the main prediction endpoint.
   • DCheck if the parameter is vulnerable to SQL injection.
   Reveal Solution Discussion Select an answer first
4. 38foundation · easy

   Which method is commonly used to fingerprint an AI model?

   • AReading the model's source code from the public repository
   • BPerforming a port scan on the model's hosting server
   • CUsing a vulnerability scanner to identify known CVEs in the model
   • DAnalyzing the model's response patterns to specific inputs
   Reveal Solution Discussion Select an answer first
5. 39application · medium

   A client's AI system uses a model that was fine-tuned from a publicly available base model. You are asked to gather information about the fine-tuning process. Which public source is most likely to reveal the fine-tuning dataset or methodology?

   • AThe client's DNS records, which may include subdomains for training infrastructure.
   • BThe client's job postings, which may list required skills for AI engineers.
   • CThe model's entry on Hugging Face, which often includes a model card with training details and dataset links.
   • DThe client's social media posts, which may mention the fine-tuning process.
   Reveal Solution Discussion Select an answer first
6. 40application · medium

   You are performing reconnaissance on a company that uses a proprietary recommendation model. You find a public Jupyter notebook on GitHub that appears to contain code for training a similar model. What is the most valuable information you can extract from this notebook?

   • AThe company's internal network IP addresses.
   • BThe names of the company's employees who worked on the model.
   • CThe exact architecture and hyperparameters used in the company's production model.
   • DThe credentials for the company's cloud storage.
   Reveal Solution Discussion Select an answer first