Use when an LLM app could be driven to exhaust resources or run up a large bill through crafted inputs — testing for unbounded consumption and the limits that contain it.
