unsloth

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

• Third-party content exposure detected (medium risk: 0.30). Exposed inputs at runtime are the user-authored fine-tuning dataset loaded via datasets.load_dataset("json", data_files="train.jsonl") and then rendered into training text by tokenizer.apply_chat_template(...), which the LLM ingests during trainer.train().