LLM Jailbreak and Prompt Injection Dataset with 691k Multilingual Examples

Necent's AI Security Dataset provides 691,331 examples for AI safety research. The collection unifies 15 source datasets covering 26 languages and various attack types. It was last updated in March 2026.

Use Cases

• Train classifiers to detect malicious intent using the `prompt_type` and `is_dangerous` labels.
• Benchmark LLM vulnerability by analyzing `response` content generated by different `model_name` targets to jailbreak `prompt` texts.
• Study attack vector patterns across `category` sub-types like obfuscation or harmful_behavior.
• Develop multilingual defense systems by leveraging the dataset's coverage of 26 languages in the `prompt` field.

Strengths

• 691,331 total examples provides a substantial corpus for model training.
• Combines 15 distinct source datasets for breadth of coverage.
• Covers 26 languages, enabling cross-lingual safety research.

Limitations

• Specific row counts, column details, and sample data are unavailable for verification.
• Potential class imbalance or label noise inherent in aggregating multiple disparate sources.
• Geographic and cultural bias may exist within the multilingual prompts.

Provenance

Source

Necent, via Hugging Face.

Collection Method

Unified aggregation of 15 existing source datasets.

Freshness

Last updated March 2026.