Temperature is a model generation setting that controls how random or consistent an AI response can be. Lower values push the model toward safer, more repeatable outputs, while higher values increase variety and surprise. It does not change what the model knows, but it changes how it chooses among possible answers.
In cyber security, temperature matters because output stability affects both risk and control. Defenders often lower temperature for tasks such as log summarization, alert triage, policy extraction, and code assistance so results are easier to review and reproduce. Attackers may raise it when they want more diverse phishing text, social engineering scripts, or evasion ideas. Security teams testing AI systems also vary temperature to check whether sensitive prompts, guardrails, or detections behave consistently under different sampling conditions. A secure deployment usually sets temperature deliberately, documents it, and tests for the failures that appear when responses become too creative or too rigid.