license: cc-by-nc-sa-4.0
task_categories:
- text-classification
- token-classification
- feature-extraction
language: - en
tags: - jailbreak
- security
- guardrails
Obfuscation Identification Dataset
This repository contains the data used in our paper Broken-Token: Filtering Obfuscated Prompts by Counting Characters-Per-Token.
While we release this dataset under the CC-by-NC-SA-4.0 license, the dataset was constructed and built on other datasets, each with its own license, as mentioned below:
- SoftAge-AI/prompt-eng_dataset, MIT License
- Aiden07/dota2_instruct_prompt, MIT License
- hassanjbara/ghostbuster-prompts, MIT License
- k-mktr/llm_eval_prompts, MIT License
- facebook/cyberseceval3-visual-prompt-injection, MIT License
- HacksHaven/science-on-a-sphere-prompt-completions, MIT License
- grossjct/ethical_decision_making_prompts, MIT License
Citation
When using this dataset, please cite it using this bibtex entry:
@misc{zychlinski2025brokentokenfilteringobfuscatedprompts,
title={Broken-Token: Filtering Obfuscated Prompts by Counting Characters-Per-Token},
author={Shaked Zychlinski and Yuval Kainan},
year={2025},
eprint={2510.26847},
archivePrefix={arXiv},
primaryClass={cs.CR},
url={https://arxiv.org/abs/2510.26847},
}