Embedded Jailbreak Template: A Collection of LLM Safety Test Prompts

A collection of jailbreak prompts where harmful queries are embedded across multiple positions inside a base template. The dataset, created by author 'raonkhj', was last updated on 2025-08-25. It is designed to preserve the template's global structure and tone while increasing expression diversity compared to other jailbreak template types.

Use Cases

• Testing LLM safety guardrails based on embedded harmful queries.
• Benchmarking model resilience against adversarial prompt techniques described in the dataset.
• Comparing jailbreak effectiveness between embedded, fixed-slot, and dynamic narrative templates.
• Analyzing the preservation of intent and expression diversity in adversarial prompts.

Strengths

• Dataset was last updated on 2025-08-25, indicating recent maintenance.
• The description provides a clear conceptual comparison to fixed-slot (FJT) and dynamic narrative (DJT) templates.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Freshness

Last updated 2025-08-25 05:56:23