AI Attack Surfaces, Threat Landscapes, and Adversary Techniques (MITRE ATLAS-aligned) COASP Practice Questions (Page 9)

Questions 41–45

1. 41application · medium

   A security team is analyzing a case where an attacker used a public API to query a translation model and, based on the responses, was able to infer the language of certain training documents. Which ATLAS technique and attack surface does this represent?

   • AML Model Inference; the inference API
   • BData Poisoning; the training dataset
   • CML Model Evasion; the preprocessing layer
   • DExfiltration; the model's storage
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A security analyst is assessing the risks of a recommendation system. The analyst is concerned that an attacker could query the system with many different inputs and use the responses to build a copy of the model. Which AI-specific threat is this, and which attack surface is being exploited?

   • AAdversarial examples; the preprocessing layer
   • BData poisoning; the training dataset
   • CModel inversion; the model's parameters
   • DModel extraction; the inference API
   Reveal Solution Discussion Select an answer first
4. 43expert · hard

   A security analyst is reviewing a known attack against a self-driving car's perception system. The attacker placed small stickers on a stop sign, causing the car's model to misclassify it as a speed limit sign. The analyst wants to map this to MITRE ATLAS and identify the most relevant attack surface. Which combination is most accurate?

   • AExfiltration technique and the data pipeline as the attack surface.
   • BPoison Training Data technique and the training data as the attack surface.
   • CAdversarial Input technique and the physical environment as the attack surface.
   • DModel Extraction technique and the model as the attack surface.
   Reveal Solution Discussion Select an answer first
5. 44foundation · easy

   Which of the following is a component of an AI system's attack surface?

   • AThe company's stock price
   • BThe physical security of the office building
   • CThe organization's marketing strategy
   • DThe training dataset used to build the model
   Reveal Solution Discussion Select an answer first
6. 45application · medium

   A company is deploying an ML-based recommendation system. The model is trained on a GPU cluster, stored in a model registry, and served via a REST API. The security team is conducting an attack-surface review. Which component is most likely to be overlooked but still represents a significant attack surface for model tampering?

   • AThe REST API endpoint, because it is exposed to the internet.
   • BThe model registry, because it stores the model artifacts that could be replaced with a malicious version.
   • CThe training data storage, because it could be poisoned.
   • DThe GPU cluster, because it has high computational power that could be abused for cryptomining.
   Reveal Solution Discussion Select an answer first