knowledge-run

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

• Third-party content exposure detected (high risk: 0.78). Outsider-authored free text can be fetched/ingested at runtime via the required phase chain—specifically knowledge-fetch (Phase 2–7) and downstream knowledge-ingest, which operate on web sources and thus may include arbitrary third-party page text that gets placed into the LLM context.