Prompting is the instruction or input used to steer a language model’s response. A prompt can be a simple question, a structured template, or a longer set of constraints that tells the model what role to play, what data to use, and how to format the output. In cyber security, prompting matters because it strongly shapes whether an AI tool produces useful analysis, safe automation, or misleading text.
Attackers can abuse prompting through prompt injection, where malicious text is placed inside emails, web pages, tickets, or documents to override the model’s intended behavior. That can push a system to reveal secrets, ignore safety rules, or take harmful actions in connected tools. Defenders use careful prompting to limit scope, separate system instructions from user content, and request evidence, citations, or structured outputs. Good prompts do not eliminate risk, but they help reduce hallucinations, policy bypass, and accidental data leakage in security workflows.