litreview

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime LLM context is populated by Consensus search results (public web/outsider-authored paper text/metadata) returned for user-supplied queries, which the agent then synthesizes into the DOCX and chat—i.e., the “Phase 1/3 Consensus-returned papers” path ingests outsider-authored content.