Product
InstructLab is vulnerable to arbitrary code execution because the `linux_train.py` script hardcodes `trust_remote_code=True` when loading models from HuggingFace, allowing remote attackers to execute code by convincing a user to load a malicious model.