Data Poisoning CT-AI Practice Questions (Page 6)

Questions 26–27

1. 26application · medium

   A medical diagnosis AI system is trained on a dataset that was accidentally poisoned. The model now has a slightly lower overall accuracy but, more critically, it misclassifies a specific rare disease as benign. What is the most significant security impact of this poisoning?

   • AThe model's confidentiality is breached because it reveals patient data
   • BThe model's integrity is compromised because it fails on a specific, high-stakes input class
   • CThe model's accuracy is only slightly reduced, so the impact is minimal
   • DThe model's availability is reduced because it is slower to make predictions
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   What is the goal of adversarial testing for data poisoning resilience?

   • ATo measure the model's accuracy on a clean test set
   • BTo deploy the model to production
   • CTo evaluate how well the model performs when exposed to intentionally poisoned data
   • DTo optimize the model's hyperparameters
   Reveal Solution Discussion Select an answer first