AI System Hacking Methodologies, Frameworks, and Risk Implications COASP Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   What is the primary purpose of the MITRE ATLAS framework in offensive AI security?

   • ATo specify a set of programming APIs for building adversarial attack tools
   • BTo provide a knowledge base of tactics and techniques specific to attacks on machine learning systems
   • CTo define a standard for encrypting AI model weights during storage and transmission
   • DTo offer a compliance checklist for AI systems under GDPR and other privacy regulations
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A large e-commerce company uses an AI model to recommend products. The model is trained on user purchase history. An attacker gains access to the training data and injects malicious records that cause the model to recommend competitor products. The company's security team is assessing the risk. Which combination of risks is MOST directly applicable?

   • AModel inversion leads to exposure of user data, and the attack may also cause the model to be stolen.
   • BModel theft leads to a loss of competitive advantage, and the attack may also cause the model to be unavailable.
   • CData poisoning leads to a loss of customer trust and revenue, and the attack may also expose the model's training data, creating a privacy risk.
   • DAdversarial manipulation leads to incorrect recommendations, and the attack may also cause the model to be retrained.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A red team is documenting an attack against a facial-recognition system used for physical access control. They want to map their tactics and techniques to a framework that is specifically designed for adversarial machine learning and that aligns with the Cyber Kill Chain. Which framework should they use?

   • AOWASP Top 10 for LLM Applications
   • BNIST Cybersecurity Framework
   • CMITRE ATLAS
   • DISO/IEC 27001
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   During the exploitation phase of an AI system attack, which action is most characteristic?

   • ACreating a malicious dataset to be used for future model retraining
   • BEstablishing command-and-control channels to maintain persistence
   • CDelivering a crafted adversarial input to the model to cause misclassification
   • DScanning the model's API endpoints for exposed metadata
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A red team is documenting an attack where they used a generative model to create deepfake audio to bypass a voice-based authentication system. They want to map this to MITRE ATLAS. Which tactic and technique combination is most appropriate?

   • ATactic: Initial Access; Technique: Valid Accounts
   • BTactic: Execution; Technique: Adversarial Input
   • CTactic: Exfiltration; Technique: Exfiltration Over C2 Channel
   • DTactic: ML Attack Staging; Technique: Create Proxy Model
   Reveal Solution Discussion Select an answer first