elimination-research

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow can ingest outsider-authored free text from public web sources (e.g., scraped pages / review/forum content) into the dataset JSON, which is then rendered into the report context (and thus into any LLM that later reads the generated artifacts), via scripts/generate_elimination_report.py → report_generator.py payload fields like evidence_summary.summary and source_data[*].notes.