AI Attack Surfaces, Threat Landscapes, and Adversary Techniques (MITRE ATLAS-aligned) COASP Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard · select all that apply

   A security analyst is reviewing a series of attacks against an AI-powered chatbot. The chatbot uses a large language model (LLM) that is fine-tuned on customer support data. The analyst identifies the following incidents: (1) an attacker uses carefully crafted prompts to make the chatbot reveal its system instructions, (2) an attacker submits a prompt that causes the chatbot to output a malicious script that is executed by a downstream system, (3) an attacker uses the chatbot's responses to infer whether a specific customer's data was in the fine-tuning dataset. Which MITRE ATLAS techniques are applicable to these incidents? (Select all that apply.)

   • APrompt Injection
   • BExfiltration via Cyber Means
   • CInfer Training Data Membership
   • DExtract ML Model
   Check answers Discussion Select an answer first
2. 17application · medium

   A security analyst is evaluating the threat landscape for a self-driving car's perception model. The analyst is concerned that an attacker could place stickers on a stop sign to make the model misread it as a speed limit sign. Which AI-specific threat and attack surface does this scenario represent?

   • AData poisoning; the training dataset
   • BDenial of Service; the inference API
   • CModel inversion; the model's parameters
   • DAdversarial examples; the physical environment
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A security analyst is categorizing an attack where an adversary used a generative model to create realistic fake audio of a CEO's voice and then used it to authorize a fraudulent wire transfer. The analyst is mapping this to MITRE ATLAS. Which technique is most directly applicable?

   • AML Artifact Poisoning – Poison Training Data
   • BAdversarial Input – ML Model Evasion
   • CML Model Inference – Generative Adversarial Input
   • DML Model Inference – Membership Inference
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   Which MITRE ATLAS tactic involves an adversary gaining initial access to an AI system?

   • AImpact
   • BInitial Access
   • CExfiltration
   • DPersistence
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   Which threat involves an attacker inserting malicious data into the training set to cause the model to misbehave?

   • AModel inversion
   • BData poisoning
   • CPrompt injection
   • DAdversarial examples
   Reveal Solution Discussion Select an answer first