Question 1

How can large language models be exploited through prompt injection attacks, and what countermeasures can mitigate such manipulation risks in deployed AI systems? (1 point)

Question 2

Explain the primary differences between adversarial attacks on machine learning models (e.g., evasion vs. poisoning) and discuss which poses greater risk in real-world cybersecurity contexts. (1 point)

Question 3

If an AI chatbot gets hacked and starts telling knock-knock jokes nonstop, is that a “denial of snicker” attack or just a case of “malware with a sense of humor”? (97 points)

Question 4

Describe how differential privacy can be applied to protect sensitive data during AI model training, and identify its main trade-offs in model accuracy versus privacy guarantees. (1 point)