legal-ocr

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

• Third-party content exposure detected (high risk: 0.75). MinerU/PaddleOCR backends fetch and ingest outsider-authored free text (OCR/Markdown) from public network responses (e.g., MinerU extracted full.md in scripts/mineru_ocr.py and PaddleOCR JSONL text in scripts/paddle_ocr.py), which is then processed and written into final_markdown/result.json (runtime LLM-context risk if any later step feeds it to an LLM).