Dataset · neuralchemy
Prompt injection and jailbreak detection dataset
neuralchemy/Prompt-injection-dataset
A balanced text set for training malicious prompt classifiers in English.
What's inside
The dataset contains from 10,000 to 100,000 English texts for binary classification: about 60% are malicious prompt injections and jailbreaks, and 40% are safe requests. Splitting into subsets is performed accounting for groups without data leakage, with two configurations available: core for classical ML and full for transformers.
What you can do with it
- Training protective filters and classifiers for detecting prompt injections
- Testing the resilience of LLM applications to jailbreak attacks
- Tasks
- Text classification
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 2.8K
- Likes
- 38
Models trained on it
- Horizon-Labs/prompt-injection-guard-baseText classification 7.4K
- Horizon-Labs/prompt-injection-guard-smallText classification 5.5K
- mradermacher/Spellguard-1.0-8B-GGUF 934
- Horizon-Labs/prompt-injection-guard-largeText classification 603
- mradermacher/GPT-Distill-Qwen3-8B-Spellguard-SLERP-GGUF 479
- thealper2/qwen3-0.6b-prompt-injection-detectorText generation 390
- neuralchemy/prompt-injection-debertaText classification 277
- mradermacher/PromptGuard-GGUF 247
How to load it
from datasets import load_dataset
ds = load_dataset("neuralchemy/Prompt-injection-dataset")