evaluating-llms-harness

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The required workflow is benchmark evaluation that ingests task prompt text (questions/contexts) into the model’s input at runtime; those task documents are sourced from external benchmark datasets/task libraries (outsider-authored free-form text), creating an indirect prompt-injection risk.