Prompt diversity is the amount of variation in the wording, structure, and intent of queries sent to an AI system. Defenders use it as a signal because legitimate users usually ask in a mixed, task-driven way, while automated extraction campaigns often generate many near-duplicate prompts with only small changes.
In cyber security, prompt diversity helps identify model-distillation abuse, output harvesting, and other forms of capability extraction. Low diversity can indicate scripted requests, repeated probing for the same behavior, or coordinated querying across multiple accounts. High diversity is not always benign, but it often makes bulk automation easier to detect when combined with other telemetry such as account reuse, request timing, and rate patterns. Security teams analyze prompt diversity alongside logs and access controls to distinguish normal experimentation from attempts to systematically learn or copy model behavior.