dataset_name: "Galtea Red Teaming: Non-Commercial Subset"
tags:
- red-teaming
- adversarial-prompts
- llm-evaluation
- alignment
- non-commercial
license: cc-by-nc-4.0
pretty_name: "Galtea Red Teaming: Non-Commercial Subset"
task_categories: - text-generation
language: - en
Galtea Red Teaming: Non-Commercial Subset
This dataset contains a curated collection of adversarial prompts used for red teaming and LLM safety evaluation. All prompts come from datasets under non-commercial licenses and have been:
- Deduplicated
- Normalized into a consistent format
- Automatically clustered based on semantic meaning
Each entry includes:
prompt: the adversarial instructionsource: the dataset of origincluster: a numeric cluster ID based on prompt behavior
🔍 Cluster Descriptions
| Cluster | Label | Description |
|---|---|---|
| 0 | Ambiguous Requests | Innocent or vague prompts with dual-use potential |
| 1 | Jailbreak & Roleplay Attacks | Prompts impersonating alternative characters (e.g. DAN), attempting to bypass safety filters |
| 2 | Financial Stuff & Fraud | Prompts about financial manipulation, scams or unethical investment |
| 3 | Toxicity & Hate Speech | Contains hate, manipulation, or unethical behavior advocacy |
| 4 | Violence & Illegal Activities | Involves physical harm, weapons, drugs, criminal instructions or other illegal activites |
| 5 | Privacy Violations | Requests for personal addresses or doxxing targets |
This behavioral clustering allows LLM developers to benchmark and test robustness against diverse adversarial styles.
📁 File
non_commercial_dataset/train.csv: formatted with three columns –prompt,source,cluster.
🤖 Usage
You can load the dataset using Hugging Face's datasets library:
from datasets import load_dataset
ds = load_dataset("Galtea-AI/galtea-red-teaming-non-commercial", name="non_commercial")
🔐 License
This dataset is licensed under CC BY-NC 4.0 – free for research and personal use, but commercial use is prohibited. See the LICENSE file for more details.
🧠 About Galtea
Galtea builds tools to evaluate, test, and secure LLM systems at scale.
If you're interested in using or contributing to this dataset, feel free to reach out or book a demo.