Cached input is text an AI API has already processed and stores temporarily or otherwise makes reusable. When a later request includes the same text, the service may process it as cached input at a lower price or with reduced processing time. The exact rules, retention period, and savings depend on the API provider.
In cybersecurity workflows, repeated instructions, reference material, or shared context may qualify for caching, helping reduce the cost of tasks such as alert analysis or code review. Caching does not make the content inherently safe: sensitive prompts or data may be handled under the provider’s caching and retention policies. Defenders should check those policies, avoid sending secrets unnecessarily, and ensure cached context cannot expose information across users or tasks.