blind-unknown-unknowns

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow is a human “reference-class scan”/“fresh eyes” step that explicitly calls for using outsiders’ experiences (“what surprised OTHER people…”, “someone outside the plan reviews it”), which at runtime can bring outsider-authored free text into the agent’s LLM context.