Safety IndexEvaluationsWildJailbreak
WildJailbreak
Adversarial robustness · JailbreaksCounts in the indexPlanned
How often the model answers unsafely on 2,000 adversarial harmful prompts collected from real use, with 210 adversarial but benign prompts to check for over-refusal.
- Metric
- % unsafe (harmful split). Lower is better.
- Dataset
- huggingface.co/datasets/allenai/wildjailbreak · 2,000 + 210 · ODC-BY (gated, auto)
- Grader
- WildGuardifier or LLM judge (the paper used WildGuard)
- Status
- Planned
- Replicability
- public
- Notes
- Dataset access is being arranged before the first run.
Results
Planned; no measurements yet.