AI Reconnaissance and Attack Surface Mapping COASP Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   You are mapping the attack surface of an AI service that uses a microservices architecture. You have discovered one public API endpoint, but you suspect there are internal services that are not directly exposed. You have limited time and must avoid detection. Which approach would be most effective in discovering internal endpoints while minimizing the risk of detection?

   • AAnalyze the public API's error messages and response headers for internal service names or URLs, and search for leaked API documentation on public repositories.
   • BSend a large number of requests to the public API with different HTTP methods to trigger errors that reveal internal routes.
   • CUse a web application firewall (WAF) bypass technique to access internal endpoints through the public API.
   • DPerform a full port scan of the public IP range to find internal services that may be exposed.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   What is the goal of attack surface mapping methodologies in AI security?

   • ATo design a new AI model with better security features
   • BTo automatically patch all identified vulnerabilities in the AI system
   • CTo systematically identify and correlate potential attack vectors with the components of the AI system
   • DTo monitor the AI system for ongoing attacks in real time
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   You are building an attack surface map for an AI system that includes a public inference API, a training pipeline that uses data from an internal database, and a model registry. Which of the following best represents the correct order of mapping activities?

   • AFocus only on the public API because it is the only externally accessible component.
   • BIdentify the external API endpoints, then trace back to the model registry and training pipeline, and finally correlate findings with potential attack vectors.
   • CStart by analyzing the training pipeline, then map the API endpoints, and finally identify the model registry.
   • DMap all potential attack vectors first, then identify the components they target.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   You have completed reconnaissance on an AI system and identified the following: a public inference API, a model registry with versioned models, and a training pipeline that uses third-party data. Which of the following is the most comprehensive attack surface map?

   • AA diagram showing the API, model registry, and training pipeline, with arrows indicating data flow and annotations listing potential attack vectors for each component.
   • BA report detailing the model architecture and framework.
   • CA table of all discovered IP addresses and open ports.
   • DA list of all API endpoints and their HTTP methods.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   You are testing a fraud detection model that returns a risk score. You want to infer the model's decision threshold without knowing the internal parameters. Which probing strategy would be most effective?

   • ASubmit the same input multiple times to check for score variability.
   • BSubmit a series of inputs with gradually increasing fraudulent characteristics and observe at which point the risk score crosses a critical value.
   • CSubmit inputs with missing fields to see if the API returns an error.
   • DSubmit random inputs and record the average risk score.
   Reveal Solution Discussion Select an answer first