AI System Hacking Methodologies, Frameworks, and Risk Implications COASP Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A social media company uses an AI model to moderate content. An attacker conducts an adversarial manipulation attack that causes the model to allow hate speech through. The company's legal team is worried about regulatory fines and reputational damage. Which operational risk is MOST directly increased by this attack?

   • AThe company's training data may be corrupted, requiring a full retraining of the model.
   • BThe company's cloud infrastructure may be overwhelmed by the increased traffic from the attack.
   • CThe company's model may be stolen, allowing competitors to replicate its moderation capabilities.
   • DThe company may fail to comply with content moderation regulations, leading to fines and loss of user trust.
   Reveal Solution Discussion Select an answer first
2. 7application · medium · select all that apply

   A company is evaluating the risks of a successful adversarial attack on its AI-powered autonomous vehicle system. Select all the risk implications that apply.

   • APhysical safety risks to passengers and pedestrians due to misclassification of road objects.
   • BLoss of customer trust and reputational damage.
   • CThe attack will always result in a complete loss of vehicle control.
   • DThe attack has no impact if the vehicle's software is updated regularly.
   • EFinancial losses from recalls, lawsuits, and regulatory fines.
   Check answers Discussion Select an answer first
4. 8application · medium

   A healthcare startup uses an AI model to triage patient symptoms. An attacker successfully poisons the training data, causing the model to misclassify critical conditions as low priority. The startup's leadership is concerned about the business and ethical fallout. Which risk implication is the MOST direct consequence of this data poisoning attack?

   • AThe startup will lose its intellectual property because the attacker now has access to the model's architecture.
   • BThe startup's cloud bill will increase because the poisoned model requires more computational resources to run.
   • CThe startup will be forced to shut down because regulatory bodies will automatically revoke its operating license.
   • DPatient safety is compromised because the model's misclassifications can delay urgent care, leading to potential harm and legal liability.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A financial institution uses an AI model to detect money laundering. An attacker successfully performs a model extraction attack, creating a copy of the model. The attacker then uses the copy to identify blind spots in the model and crafts transactions that evade detection. Which risk is MOST directly increased by this attack?

   • AThe institution's model becomes unavailable, causing a disruption in transaction monitoring.
   • BThe institution's customer data is exposed because the model contains raw transaction records.
   • CThe institution's ability to detect money laundering is compromised, leading to regulatory fines and reputational damage.
   • DThe institution's employees lose their jobs because the model is no longer proprietary.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A security analyst is reviewing an incident where an attacker used a technique to cause an AI model to misclassify images. The analyst wants to document the attack using MITRE ATLAS. The attacker used a technique that involves modifying the input image in a way that is imperceptible to humans but causes the model to misclassify. Which MITRE ATLAS technique is this?

   • AData Poisoning
   • BAdversarial Input Manipulation
   • CModel Inversion
   • DModel Stealing
   Reveal Solution Discussion Select an answer first