API Reconnaissance and Model Fingerprinting COASP Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   Which of the following is a sign of training data leakage in an AI API's response?

   • AThe API uses HTTP/2
   • BThe API response includes a timestamp
   • CThe API returns a 404 error for unknown endpoints
   • DThe model returns exact snippets of text that appear to be from its training data
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Which test would BEST reveal an authorization flaw in an API?

   • ASending malformed JSON to an endpoint
   • BChecking for the presence of security headers
   • CMeasuring response time for different endpoints
   • DAccessing a resource with a low-privileged user's token
   Reveal Solution Discussion Select an answer first
4. 28application · medium

   While testing a chatbot API, you send a malformed request that causes the API to return a verbose error message. The error message includes a stack trace that reveals the underlying ML framework (e.g., TensorFlow) and the model file path. What is the most appropriate next step?

   • AIgnore the error message and continue with other tests.
   • BTry to access the model file path directly via the web server.
   • CImmediately report the error message to the client and stop testing.
   • DUse the framework and path information to search for other endpoints that might expose the model file or additional metadata.
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   You are tasked with extracting a copy of a sentiment analysis model hosted behind an API. The API has a rate limit of 100 requests per minute. You have a budget of 10,000 queries. What is the most efficient strategy to extract the model's functionality within these constraints?

   • AUse a pre-trained model as a surrogate and only query the API to fine-tune it on a few examples.
   • BSend 10,000 random inputs and record the outputs to train a surrogate model.
   • CUse active learning to select the most informative inputs, such as those near the decision boundary, and query them in batches to stay within the rate limit.
   • DSend all 10,000 queries as fast as possible to avoid the rate limit resetting.
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A security engineer is testing an AI-based email spam filter API for robustness. The engineer wants to generate adversarial emails that bypass the filter while remaining readable to humans. The API has a rate limit and returns a spam probability score. Which technique is most suitable for this task?

   • AUse a genetic algorithm to evolve email texts that minimize the spam probability score while preserving a high similarity to a benign email.
   • BUse a pre-trained adversarial attack library to generate perturbations and test them against the API.
   • CManually craft emails with common spam-filter bypass techniques such as homoglyphs and word spacing.
   • DApply a gradient-based attack using the API's confidence score as a loss function to craft adversarial emails.
   Reveal Solution Discussion Select an answer first