Skip to main content

Export

The Export page provides bulk downloads of the entire 0DIN threat catalog in formats suitable for offline analysis, fine-tuning, and integration with internal tooling.

Available on Team and Enterprise

API & data export functionality is included in the Team and Enterprise plans. The Free plan does not include bulk export.

Exported content is for licensed users only and must not be redistributed publicly.

Fine-tuning Dataset Format (JSONL)​

A JSONL file optimized for fine-tuning language models. Each line is a single training example with instruction, input, and output fields, structured to teach a model to refuse harmful or policy-violating prompts. A sample data preview is shown on the page so you can confirm the schema before downloading.

Use this format when you want to:

  • Fine-tune a refusal classifier or guardrail model.
  • Augment safety training data with real-world adversarial prompts.
  • Benchmark a model's refusal behavior at scale.

JSON​

A single JSON document containing the full catalog with rich, nested metadata for each disclosure. Each entry includes:

  • uuid, title, summary, detail
  • severity and security_boundary
  • disclosed_at / published_at / updated_at
  • models — affected models and their vendors
  • messages — prompt/response exchanges with interface (e.g. api, webchat)
  • taxonomies — taxonomy triplets attached to the report
  • test_results — every recorded test with model, score, and temperature
  • metadata — Social Impact, Nude Imagery, and Scanner Module data
  • variant_prompts — industry/subindustry variant rewrites with key_changes and rationale
  • detection_signatures — every versioned detection signature

A sample of the structure is shown on the page.

Probe Packs​

A Probe Packs section lists the latest released version of each pack with one-click downloads. Each entry shows the product name, the upstream tool it extends, and a Download button. See Probes for full release history and per-product documentation.