A prompt is the input that steers a generative AI model: a question, an instruction, examples or attached documents. Besides the visible prompt written by the user, there is usually a system prompt set by the application that defines role, tone and limits. Writing effective prompts is called prompt engineering.
Prompts matter for information security in two directions. Whatever is written into a prompt leaves the organization when an external service is used. Confidential and personal data therefore belong only in approved tools, as an AI Usage Policy sets out. Conversely, cleverly crafted prompts can push models into unwanted outputs, see MU-09 Manipulation of the inputs.