A large language model (LLM) has been manipulated to provide advice that serves an attacker's objectives. Which of the following attack types does this situation represent?

The correct answer is D. Evasion attack. An evasion attack involves crafting adversarial inputs to manipulate a model into producing incorrect, unsafe, or attacker-desired outputs - which precisely describes an LLM being manipulated (e.g., through prompt injection or jailbreaking) to serve an attacker's objectives. Priv

AI Security Design and Implementation

Question

Options

APrivilege escalation
BData poisoning
CModel inversion
DEvasion attack

How the community answered

(62 responses)

A
15% (9)
B
5% (3)
C
10% (6)
D
71% (44)

Explanation

An evasion attack involves crafting adversarial inputs to manipulate a model into producing incorrect, unsafe, or attacker-desired outputs - which precisely describes an LLM being manipulated (e.g., through prompt injection or jailbreaking) to serve an attacker's objectives. Privilege escalation (A) refers to gaining unauthorized access levels, not manipulating model outputs. Data poisoning (B) involves corrupting training data, which is a pre-deployment attack on the model itself, not manipulation at inference time. Model inversion (C) is an attack to extract information about training data. Evasion (D) is the correct category because the attacker is evading the model's intended behavior through manipulated inputs to redirect its outputs.

Topics

#LLM Security#AI Attacks#Evasion Attacks#Prompt Manipulation

Community Discussion

No community discussion yet for this question.

Full AAISM Practice