fine-tuning-expert

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

• Third-party content exposure detected (medium risk: 0.30). The workflow’s training pipeline reads user-provided free-text from a dataset file at load_dataset("json", data_files={"train": "train.jsonl", "test": "test.jsonl"}) / load_custom_dataset(raw_data_path) and then formats and tokenizes fields like instruction, input, output, or messages[*].content for quality validation and training.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). references/deployment-optimization.md contains calls to AutoModelForCausalLM.from_pretrained(..., trust_remote_code=True) which will fetch and execute remote repository code for the model named "meta-llama/Llama-3.1-8B" at runtime, enabling execution of remote code from that external hub repo.