media-ocr-ai

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text can enter the LLM context via the OCR outputs themselves: at runtime scripts/ocr.py reads user-supplied images (--in) and prints extracted text/JSON/TSV (cmd_extract, cmd_layout, cmd_table, cmd_handwriting), which the agent may then include in its LLM context; since the image content is not authored by the operating user, it is outsider-authored text.