Prompt classification is the process of sorting AI requests into categories based on their content, intent, risk, or required handling. A system might label a prompt as benign, sensitive, administrative, code-related, or potentially malicious before the request is sent to a model or tool. This is useful because not every AI interaction should be treated the same way: a casual question, a request that includes customer data, and a prompt asking an agent to run commands all carry different security and compliance implications.
In cyber security, prompt classification helps defenders monitor AI usage, enforce policy, and detect unusual behavior. It can trigger stricter controls for sensitive prompts, block requests that look like data exfiltration or prompt injection, and flag bursts of risky activity from a single account. Attackers may try to hide malicious instructions inside normal-looking text, so classification is often paired with identity checks, logging, and content inspection. Used well, it turns AI traffic into an observable, governable workflow instead of an opaque channel.