data-labeling-qa

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

• Third-party content exposure detected (high risk: 0.80). The demo explicitly downloads a public dataset via load_dataset("ag_news") in scripts/demo.py and then sends each row's "data" text to the LLM judge prompt (run_judge cell), so open/public third‑party content is ingested and directly influences relabeling and downstream actions.