paddleocr

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is the OCR’d content from the user-supplied input document (e.g., invoice/contract image/PDF) which is then concatenated into full_text and embedded into the LLM prompt in scripts/doc_pipeline.py via llm_summarize/llm_answer/llm_detect_type/llm_check_anomalies (DOCUMENT: {full_text[:...]}).