fact-checking

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

• Third-party content exposure detected (high risk: 0.70). The required workflow reads user-provided paper file(s) (outsider content to the agent if the operating user didn’t author them), extracts verbatim claim text, and then includes that extracted free text in the agents’ prompts and thus the LLM context (in “Files to read” / claim text injection into Phase 2/3 and later reconciliation).