AI System Hacking Methodologies, Frameworks, and Risk Implications COASP Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A security team is using MITRE ATLAS to categorize an attack where an adversary exploited a vulnerability in a machine learning model's deserialization process to execute arbitrary code. Which ATLAS tactic does this technique fall under?

   • AInitial Access
   • BExfiltration
   • CML Attack Staging
   • DImpact
   Reveal Solution Discussion Select an answer first
2. 32expert · hard

   A government agency uses an AI model to screen visa applications. An attacker conducts a data poisoning attack that causes the model to approve applicants who should be rejected. The agency is concerned about the national security implications. Which risk is MOST critical?

   • AThe attack may cause the agency to lose public trust, leading to political fallout.
   • BThe attack may allow dangerous individuals to enter the country, posing a direct threat to national security.
   • CThe attack may expose the personal data of visa applicants, leading to privacy violations.
   • DThe attack may cause the model to be shut down, leading to a backlog of visa applications.
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A security analyst is planning an offensive assessment of an AI-based recommendation system. The analyst wants to follow a systematic hacking methodology that includes reconnaissance, weaponization, and exploitation. During the weaponization phase, which activity is MOST appropriate?

   • AEvaluating the business impact of a successful attack on the model.
   • BCrafting a malicious input that will be delivered to the model during the exploitation phase.
   • CDeploying a backdoor into the model's training pipeline.
   • DScanning the network to identify the model's hosting infrastructure.
   Reveal Solution Discussion Select an answer first
5. 34application · medium

   A financial institution uses an AI model to approve loan applications. An attacker subtly manipulates the training data so that applicants from a specific demographic are systematically denied loans. Which type of AI attack is this, and what is the most significant business risk?

   • AModel theft; the risk is loss of intellectual property and competitive advantage.
   • BData poisoning; the risk is regulatory fines and reputational damage from biased decisions.
   • CModel inversion; the risk is exposure of sensitive training data.
   • DAdversarial evasion; the risk is that attackers bypass the model's fraud detection.
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A cybersecurity team is documenting an attack on their AI chatbot. They want to classify the attack using MITRE ATLAS to improve their defensive posture. The attack involved an attacker sending specially crafted messages that caused the chatbot to reveal sensitive information from its training data. Which MITRE ATLAS tactic does this attack primarily fall under?

   • AImpact, because the attack caused a denial of service.
   • BPersistence, because the attacker maintained a presence in the system.
   • CExfiltration, because the attacker extracted confidential data from the model.
   • DInitial Access, because the attacker gained access to the chatbot's underlying system.
   Reveal Solution Discussion Select an answer first