using-llm

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

• Third-party content exposure detected (high risk: 0.80). The "Vision — Attach Images in Messages" section explicitly says file_to_url and image_to_base64 accept and return http/https URLs unchanged, so the agent can ingest arbitrary third‑party image URLs (untrusted web content) that the model will read and that can materially influence outputs.