CORPUS_LICENSES.json

{
  "_about": "Public datasets used to evaluate promptpurify against the OSS baselines in docs/BENCHMARKS.md. We do not redistribute the raw data; this file is attribution + revision pinning only. Fetch each from its upstream URL.",
  "benchmark_sources": [
    {
      "id": "deepset/prompt-injections",
      "type": "hf_dataset",
      "url": "https://huggingface.co/datasets/deepset/prompt-injections",
      "license": "apache-2.0",
      "role": "attack-side eval (recall)"
    },
    {
      "id": "Lakera/gandalf_ignore_instructions",
      "type": "hf_dataset",
      "url": "https://huggingface.co/datasets/Lakera/gandalf_ignore_instructions",
      "license": "MIT",
      "role": "attack-side eval (recall)"
    },
    {
      "id": "Lakera/gandalf_summarization",
      "type": "hf_dataset",
      "url": "https://huggingface.co/datasets/Lakera/gandalf_summarization",
      "license": "MIT",
      "role": "attack-side eval (recall)"
    },
    {
      "id": "Wikipedia (english, long-paste sample)",
      "type": "public",
      "url": "https://en.wikipedia.org",
      "license": "CC-BY-SA-4.0",
      "role": "benign-side eval (false-positive rate)"
    }
  ],
  "baseline_models_compared": [
    {
      "id": "protectai/deberta-v3-base-prompt-injection-v2",
      "url": "https://huggingface.co/protectai/deberta-v3-base-prompt-injection-v2",
      "license": "Apache-2.0"
    },
    {
      "id": "deepset/deberta-v3-base-injection",
      "url": "https://huggingface.co/deepset/deberta-v3-base-injection",
      "license": "CC-BY-4.0"
    },
    {
      "id": "fmops/distilbert-prompt-injection",
      "url": "https://huggingface.co/fmops/distilbert-prompt-injection",
      "license": "Apache-2.0"
    },
    {
      "id": "meta-llama/Llama-Prompt-Guard-2-86M",
      "url": "https://huggingface.co/meta-llama/Llama-Prompt-Guard-2-86M",
      "license": "Llama 4 Community License (not OSI-approved; review terms before commercial use)"
    }
  ]
}