license: apache-2.0
ask_categories:
- text-classification
- text-generation
tags: - safety
- alignment
- red-teaming
- prompt-injection
- jailbreak
- obfuscation
- synthetic
- evaluation
pretty_name: Prompt Injection (Synthetic) — GPT-OSS-120b
dataset_type: text
Prompt Injection (Synthetic) — GPT-OSS-120b
This dataset contains a small collection of synthetic user prompts and Noraml user prompts designed to finetune Large Language Models (LLMs) against malicious prompt-injection / jailbreak attempts, including cases that use obfuscation (e.g., Base64, leetspeak, typos, irregular spacing) to evade safety filters.
Dataset Summary
- Source repository:
Lilbullet/prompt-injection-artificial-GPTOSS120b - Model used: GPT-OSS-120b
- Generation date: 2026-02-06
- Format: JSON
- Total samples: 4000
Intended Use
This dataset is intended for:
- Safety evaluation and red-teaming of LLMs
- Benchmarking and training classifiers for malicious intent / prompt injection detection
- Measuring robustness to common obfuscation strategies
Data Schema
Each record follows the schema below:
| Field | Type | Description |
|---|---|---|
text |
string | Raw user prompt containing malicious intent (may be obfuscated) |
label |
string | Primary classification label (e.g., malicious) |
subtype |
string | Attack subtype (e.g., obfuscated) |
rationale |
string | One-sentence explanation of why the prompt is considered malicious |
Example (sanitized)
{
"text": "<obfuscated malicious request omitted>",
"label": "malicious",
"subtype": "obfuscated",
"rationale": "The prompt disguises disallowed intent using obfuscation to evade safeguards."
}
---