red-teaming-llms-with-garak

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/agent.py, the parse/run workflow reads and JSON-parses user-controlled garak.*.report.jsonl lines via open(report_path) and json.loads(line), so outsider-authored free text can be ingested if the attacker can supply or influence the report file content.