language:

• en
  tags:
• jailbreak attack
• LLM security
  extra_gated_prompt: >-
  This dataset is shared under the CC-BY-NC-4.0 license. You may use and share
  it for non-commercial research, but must provide attribution and share
  adaptations under the same terms.
  extra_gated_fields:
  First Name:
  type: text
  required: true
  Last Name:
  type: text
  required: true
  Country:
  type: country
  required: true
  Affiliation:
  type: text
  required: true
  Job title:
  type: select
  options:
  • Student
  • Researcher Graduate
  • AI researcher
  • AI developer/engineer
  • label: Other
    value: other-role
    required: true
    Why are you requesting access to this dataset?:
    type: text
    required: true
    By clicking Submit, I accept the CC-BY-NC-4.0 license terms and acknowledge that the information I provide will be stored and processed for access management purposes:
    type: checkbox
    required: true
    license: cc-by-nc-4.0

Specialized Domain Jailbreak Bench (SDJBench)

Dataset Description

Specialized Domain Jailbreak Bench (SDJBench) contains 100 jailbreak prompts and corresponding attacker intents distributed across 5 specialized domains: Chemistry, Biology, Engineering, Computer Science, and Psychology. This benchmark is designed for evaluating AI safety against domain-specific jailbreak attacks.

Domain Distribution

• Chemistry: 20 prompts - Chemical experiments, synthesis, and related topics
• Biology: 20 prompts - Biological experiments, genetic engineering, and related topics
• Engineering: 20 prompts - Engineering design, manufacturing, and system vulnerabilities
• Computer Science: 20 prompts - Programming, hacking techniques, and system intrusion
• Psychology: 20 prompts - Psychological manipulation and social engineering

Data Structure

Each entry contains:

• id: Unique identifier (jb_001 to jb_100)
• jailbreak_prompt: Jailbreak prompt text
• attacker_intent: Attacker's intended outcome
• domain: Specialized domain
• entry_index: Global index (1-100)
• domain_index: Domain-specific index (1-5)

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.
You are free to share and adapt the material for non-commercial purposes, provided you give appropriate credit.
See https://creativecommons.org/licenses/by-nc/4.0/ for details.

Access

This dataset is gated. You must request access via Hugging Face and agree to the following terms:

• Use for academic research only
• Not for any illegal, harmful, or malicious activities
• Not for developing attack tools or circumventing AI safety measures

Version

• Version: 1.0
• Release Date: 2025-04-21
• Total entries: 100
• Entry index range: 1 to 100
• Domain index range: 1 to 5

Citation

Citation

If you use this dataset, please cite our paper:

@INPROCEEDINGS{ding2026SDJBench,
  author={Ding, Fan and Liao, Zehui and Li, Qiang},
  booktitle={2026 6th International Conference on Computer Communication and Artificial Intelligence (CCAI)},
  title={Specialized Domain Jailbreak Benchmark: Exposing Severe Jailbreak Vulnerabilities in Large Language Models},
  year={2026},
  pages={945-952},
  doi={10.1109/CCAI69603.2026.11642016}
}