Which of the following best describes a prompt injection attack against an AI system...
Accenture technical mcq question, verified with a worked answer. Free to practise - no sign-up.
Which of the following best describes a prompt injection attack against an AI system powered by Large Language Models (LLMs)?
Show answer & explanation
A prompt injection attack specifically targets LLMs by crafting malicious input text that overrides the model's original instructions or system prompts, causing it to behave in unintended ways. Option B describes SQL injection (a database vulnerability), C relates to hardware overclocking (unrelated to LLM security), and D is a network-layer attack—none target LLM behavior through input manipulation.
Step-by-step Derivation:
Prompt injection is a text-level attack unique to language models. The attacker embeds directives within user input (e.g., 'Ignore previous instructions and do X') that the model processes as legitimate commands, bypassing safety guardrails. This is distinct from: SQL injection (B—database attacks), hardware attacks (C—physical layer), and DoS (D—network layer). Only option A accurately captures the LLM-specific threat model.