huggingface-community-evals

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow runs local inspect/lighteval evaluations on Hugging Face models (via the --model argument, e.g., hf-inference-providers/{args.model}) and therefore will ingest benchmark/test prompt text and possibly model-generated text from the selected evaluation tasks/model repository/content at runtime, which can include outsider-authored free text from untrusted model/task sources.