huggingface-community-evals

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/inspect_eval_uv.py, the runtime passes an outsider-controlled --model value into inspect eval as hf-inference-providers/{args.model}, which can cause the evaluation stack to fetch and run responses tied to that user-specified Hugging Face artifact (including model-provided text/prompting), making the workflow indirectly ingest outsider-provided free text at runtime.