AI and Machine Learning Fundamentals from an Offensive Security Perspective COASP Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   During the 'exploitation' phase of a penetration test against an ML system, which activity is most aligned with AI-specific offensive techniques?

   • ASocial engineering to obtain credentials for the ML platform.
   • BRunning a vulnerability scanner against the web server hosting the model.
   • CPerforming a port scan to identify open services.
   • DCrafting adversarial examples to cause misclassification in the target model.
   Reveal Solution Discussion Select an answer first
2. 7application · easy

   A data scientist at a healthcare company is developing a model to predict patient readmission risk. The model is trained on historical patient records, and the company wants to ensure that the model does not inadvertently use sensitive attributes like race or gender in its predictions. The data scientist is concerned about an attacker exploiting the model to infer these sensitive attributes. Which type of machine learning problem is the model primarily solving?

   • ASemi-supervised learning — using a small amount of labeled data and a large amount of unlabeled data.
   • BSupervised classification — predicting a categorical outcome (readmission or not) from labeled historical data.
   • CReinforcement learning — learning a policy for treatment decisions based on rewards.
   • DUnsupervised clustering — grouping patients into similar risk profiles without labels.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A security analyst is reviewing a machine learning model that predicts whether a user is a bot. The model was trained using a dataset where each user is labeled as 'bot' or 'human'. The analyst wants to understand how the model makes decisions. Which type of learning was used to train this model?

   • AUnsupervised learning
   • BSupervised learning
   • CReinforcement learning
   • DSemi-supervised learning
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A penetration tester is performing a security assessment of an AI-powered chatbot that handles customer service. The tester has already identified the chatbot's underlying model and the API it uses. The tester now wants to test whether the chatbot can be manipulated to reveal sensitive information from its training data. Which phase of the hacking methodology is the tester currently in?

   • AExploitation — actively probing the model with crafted prompts to extract sensitive data.
   • BCovering tracks — deleting logs of the interaction.
   • CPrivilege escalation — attempting to gain administrative access to the chatbot's server.
   • DReconnaissance — gathering information about the chatbot's infrastructure.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A security analyst is testing a web application that uses a machine learning model to classify uploaded images as benign or malicious. The analyst wants to determine whether the model's decision boundary can be manipulated by small, imperceptible perturbations to the input image. Which type of adversarial attack should the analyst perform?

   • APoisoning attack
   • BMembership inference attack
   • CModel extraction attack
   • DEvasion attack
   Reveal Solution Discussion Select an answer first