adversarial-legal-review-en

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

• Third-party content exposure detected (high risk: 0.75). The workflow explicitly includes an “AUGMENTED” attacker/citation-checker step that “pull[s] real sources for the relevant jurisdiction” and “see[s] the sources + intake,” meaning runtime-fetched public web content (outsider-authored) can be ingested as readable text into the LLM context for grounding/contrary authority.