nature-writing

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

• Third-party content exposure detected (medium risk: 0.30). The runtime routing/drafting workflow ingests outsider-authored free text provided by the user (manuscript paragraphs/notes/claims/evidence) and then processes it through its conditional routing and drafting steps (drafting core/workflow and the “polish-existing” path when requested), so the LLM reads non-first-party text at runtime.