An AI prompt is the text, instructions, or context given to an AI system to shape its output, analysis, or behavior. In cybersecurity, prompts are used to draft reports, summarize logs, map controls, or help analysts investigate incidents. The quality of the prompt strongly affects the quality of the result: clear scope, trusted source data, and explicit constraints usually produce safer, more useful outputs.
Prompts matter because they can also become an attack surface. In prompt injection, a malicious document, email, or web page hides instructions that steer the model away from the user’s intent. Defenders reduce this risk by separating trusted instructions from untrusted content, limiting what data the model can see, validating outputs before use, and keeping humans responsible for final decisions. In practice, a prompt is not just a query; it is part of the security control around an AI workflow.