Misinformation refers to LLMs producing false or misleading information that seems credible due to hallucinations.

• A. Prompt Injection
• B. Sensitive information disclosure
• C. Data and model poisoning
• D. Misinformation

Answer: (D)

When a language model generates content that sounds credible but is false or misleading, it’s describing misinformation. This happens through hallucinations, where the model fills in gaps with plausible-sounding details that aren’t based on factual data. The key idea is the discrepancy between how convincing the output appears and how accurate it actually is.

This is different from prompt injection, which is about manipulating inputs to force the model to behave in an unwanted way; it’s a method of interaction, not a description of the factual quality of the content. It’s also different from sensitive information disclosure, which is about leaking private or restricted data, and from data or model poisoning, which involves tampering with the training data or the model to cause biased or incorrect behavior. The scenario described specifically targets the generation of false or misleading content that users might take as true, which is why it fits as misinformation.