llm-eval-lab

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

• Third-party content exposure detected (medium risk: 0.30). In grid_eval.py the grid-eval runtime loads question input/expected text from a user-supplied ground-truth JSON file (--ground-truth) and then sends that free text into LLM judging (_agent_judge) and model calls via call_provider without selecting a specific pre-approved item first.