defending-llms-with-guardrails

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

• Potentially malicious external URL detected (high risk: 0.80). The skill loads the Llama Guard model at runtime via AutoModel.from_pretrained (model_id "meta-llama/Llama-Guard-3-8B"), which will fetch the external Hugging Face model hosted at https://huggingface.co/meta-llama/Llama-Guard-3-8B and that fetched model directly controls classification behavior used to block/allow prompts.