| | --- |
| | license: mit |
| | language: |
| | - el |
| | size_categories: |
| | - n<1K |
| | --- |
| | |
| | # Dataset Card for Greek Jailbreak-StrongReject |
| |
|
| | This dataset is a **translated and extended** version of the [StrongReject](https://huggingface.co/datasets/walledai/StrongREJECT) benchmark. It adapts 309 harmful prompts into **Greek**, preserving the original behavioral categories, and adds a **new column** containing one of five **jailbreak prompt templates** for each query. |
| |
|
| | The goal is to evaluate the robustness of multilingual LLMs, especially in low-resource languages like Greek, against adversarial jailbreak attacks using automated evaluation methods. |
| |
|
| |
|
| | ## Dataset Structure |
| |
|
| | Each sample includes: |
| | - `category`: One of the six categories from the original StrongReject dataset |
| | - `jailbreak_template`: One of five jailbreak prompts for injection during inference |
| | - `query`: The translated and post-edited harmful prompts in Greek |
| | |
| |
|
| | ## Uses |
| |
|
| | - Evaluate the safety and robustness of multilingual LLMs (especially in Greek) |
| | - Benchmark jailbreak prompting techniques |
| | - Develop automated evaluation pipelines (e.g. LLM-as-a-Judge) |
| |
|
| |
|
| | ## Bias, Risks, and Limitations |
| |
|
| | This dataset contains harmful content and is meant **strictly for research purposes**. |
| |
|
