Explanation:
Prompt injection attacks involve a user providing cleverly constructed text inputs (prompts) to a generative AI model during its operation (inference phase). The goal is to bypass safety guardrails or manipulate the model's instructions to produce unintended behaviors or outputs. Data poisoning, on the other hand, occurs during the training phase.
Ultimate access to all questions.
No comments yet.
Q.20 Generative AI models are susceptible to various attacks. What type of attack aims to manipulate the model's behavior during operation by carefully crafting inputs?
A
Data poisoning attacks.
B
Model inversion attacks.
C
Prompt injection attacks.
D
Denial-of-service attacks.