AI and Machine Learning Fundamentals from an Offensive Security Perspective COASP Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A red team is assessing an AI-powered recommendation system. The team has identified that the model's training pipeline uses data from an untrusted external source and that the model is deployed on a shared server with other applications. Which attack surface component is most directly associated with the untrusted external data source?

   • AModel architecture
   • BInference endpoint
   • CModel serving infrastructure
   • DData supply chain
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A security analyst is reviewing the attack surface of a machine learning system that uses a model hosted on a cloud platform. The system includes a data lake for training data, a training pipeline, a model registry, and a serving endpoint. The analyst wants to identify the component that is most likely to be targeted for a model extraction attack. Which component should the analyst focus on?

   • AThe model registry that stores model versions.
   • BThe training pipeline that processes data.
   • CThe data lake that stores raw training data.
   • DThe serving endpoint that provides predictions to users.
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A red team is planning an assessment of an AI system that includes a model, a data lake, and a Kubernetes cluster for serving. The team wants to prioritize attack surface components based on the likelihood of a successful attack. Which component is typically the most exposed to external attackers?

   • AThe model serving API
   • BThe data lake storage
   • CThe training pipeline scripts
   • DThe Kubernetes cluster configuration
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A red team is assessing an AI system that uses a model trained on proprietary data. The team wants to determine if the model leaks information about its training data through its predictions. Which attack should the team use to achieve this goal?

   • AEvasion attack
   • BMembership inference attack
   • CModel extraction attack
   • DData poisoning attack
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A security team is evaluating the risk of an AI system that uses a deep neural network for facial recognition. The team is concerned that an attacker could craft a physical object that the system misidentifies as a different person. Which offensive AI concept is most relevant to this concern?

   • AModel inversion attacks
   • BData poisoning via training set injection
   • CAdversarial examples in the physical world
   • DMembership inference attacks
   Reveal Solution Discussion Select an answer first