Dataset · TrustAIRLab
In-the-wild jailbreak prompts dataset
TrustAIRLab/in-the-wild-jailbreak-prompts
15,140 jailbreak prompts collected from December 2022 to December 2023.
What's inside
The dataset contains 15,140 prompts for bypassing LLM restrictions, collected from real-world sources between December 2022 and December 2023. It is the official repository for the ACM CCS 2024 paper, gathered via the JailbreakHub framework. Tabular and text modalities are specified, with a text generation task.
What you can do with it
- Evaluating LLM robustness against jailbreaks
- Fine-tuning classifiers or filters for harmful prompts
- Testing red-team systems
- Tasks
- Text generation
- Size
- 10K–100K rows
- Downloads this month
- 3.7K
- Likes
- 71
Models trained on it
- Horizon-Labs/prompt-injection-guard-baseText classification 7.4K
- Horizon-Labs/prompt-injection-guard-smallText classification 5.5K
- Horizon-Labs/prompt-injection-guard-largeText classification 603
- shesettipavankumarswamy/ahd-jailbreak-v1Text classification 129
- vllm-sr/Decision-1.0-Route-0.6BZero shot classification 103
- yzdvai/tiny-injection-guardText classification 97
- gantrly/fortress-xsText classification 64
- vllm-sr/toolcall-sentinelText classification 52
How to load it
from datasets import load_dataset
ds = load_dataset("TrustAIRLab/in-the-wild-jailbreak-prompts")