research-paper-review

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

• Third-party content exposure detected (high risk: 0.75). The required workflow ingests the user-provided paper content (PDF/LaTeX/plain text) via the agent’s “Read the Paper” step; if that paper text is authored by outsiders (i.e., not the operating user), it becomes LLM-readable prose in the agent context, creating indirect prompt-injection risk from the paper’s body text.