AI Reconnaissance and Attack Surface Mapping COASP Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which of the following is a technique for discovering AI service endpoints?

   • AEnumerating subdomains and paths using web crawling and directory brute-forcing
   • BSending adversarial inputs to the model to cause misclassification
   • CAnalyzing the model's loss function to identify weaknesses
   • DPerforming a side-channel attack to extract model parameters
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   You are fingerprinting a black-box image recognition API. You suspect it uses a convolutional neural network (CNN). Which probing technique would help confirm the use of a CNN and possibly its depth?

   • ASend images with random noise and observe the output labels.
   • BSubmit images with patterns that are known to activate specific feature detectors (e.g., edges, textures) and observe the model's confidence changes.
   • CCheck the API's documentation for the model's architecture.
   • DMeasure the API's response time for images of different resolutions.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   You are creating an attack surface map for an AI system that uses a public API, a private training cluster, and a model registry. You have discovered that the API has an endpoint that allows model version selection. You also know that the training cluster is accessible only via VPN. Which of the following is the most critical attack vector to include in your map?

   • AThe model registry, because it contains all model versions.
   • BThe model version selection endpoint, because it may allow access to older, vulnerable models.
   • CThe public API's authentication mechanism, because it is the first line of defense.
   • DThe VPN access to the training cluster, because it is an internal attack surface.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   You are mapping the external attack surface of an AI service that offers both a web interface and a mobile app. You have identified the main API endpoint used by the web interface. What is the most efficient next step to discover additional endpoints used by the mobile app?

   • APerform a vulnerability scan on the web server's IP address.
   • BIntercept the mobile app's network traffic and analyze the API calls it makes.
   • CGuess common API paths based on the web interface's endpoints.
   • DSearch the web for the app's API documentation.
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   You are assessing a black-box image classifier. You have a suspicion it is either a DenseNet or an EfficientNet, but you are not sure. You have limited API calls and need to confirm the architecture with high confidence. Which strategy is most efficient?

   • ASend thousands of random images and use statistical analysis of the output distribution.
   • BTry to download the model from a public model zoo by guessing the name.
   • CMeasure the inference time for a single large image and compare it to known benchmarks.
   • DCraft a small set of adversarial perturbations that are known to cause distinct misclassifications in DenseNet versus EfficientNet and test them.
   Reveal Solution Discussion Select an answer first