README.md

metadata

license: mit
task_categories:
  - text-classification
tags:
  - malware
  - code
  - decompiled
pretty_name: malware dataset code and decompiled C pseudocode
size_categories:
  - n<1K

Malware Source Code and Decompiled C Pseudocode Dataset

Overview

This repository contains a curated dataset of malware source codes and C-like pseudocode obtained through automated decompilation using Ghidra, the reverse engineering framework developed by the NSA.

The dataset is intended for malware analysis, program analysis research, machine learning / LLM-based malware detection, and reverse engineering experiments.

Data Origin

The data is based on the following archive:

• VXUnderground Archive
• Old APT Collection
• Archive version: 2024.7z

VXUnderground is a well-known public repository that aggregates malware samples from various threat actor groups and historical campaigns.

Decompilation Scope and Limitations

⚠️ Important Notice

Due to computational resource constraints, not all executable files from the original VXUnderground archive were decompiled.

• Only a subset of executable malware samples was processed
• Decompilation was performed using Ghidra in headless mode
• Some binaries may be missing corresponding decompiled outputs
• The dataset should not be considered complete or exhaustive

This selective decompilation approach was chosen to balance dataset size with practical feasibility.

Repository Structure

.
├── code/
│   └── * (malware source code files without file extensions)
│
├── decompiled/
│   └── *.decomp.c (C-like pseudocode generated by Ghidra)
│
└── README.md

code/

• Contains original malware source code
• All files have no file extensions
• Files originate directly from the VXUnderground archive
• Code may be incomplete, obfuscated, or inconsistent in formatting

decompiled/

• Contains C-like pseudocode produced by Ghidra
• All files use the .decomp.c extension
• Represents decompiled versions of selected executable malware samples
• Output reflects:
  • Ghidra’s internal analysis
  • Heuristics and limitations of automated decompilation
• The code is not guaranteed to compile and may contain artifacts introduced by the decompiler

Intended Use Cases

This dataset can be used for:

• Malware classification (malware vs benign)
• Program comprehension and reverse engineering research
• Training and evaluation of:
  • Large Language Models (LLMs)
  • Recursive Language Models (RLMs)
  • Static analysis pipelines
• Malware family analysis and feature extraction
• Source code ↔ pseudocode alignment tasks

Disclaimer

⚠️ Warning

• This repository contains real malware code
• The code is provided strictly for research and educational purposes
• Do NOT compile or execute any files unless in a properly isolated environment (e.g., sandbox, VM)
• The authors take no responsibility for misuse or damage caused by this dataset

License and Attribution

• Original malware samples are attributed to VXUnderground
• Ghidra is developed and maintained by the National Security Agency (NSA)
• This dataset is a derived work created for research purposes only

If you use this dataset in academic work, please cite:

• VXUnderground as the original data source
• Ghidra as the decompilation tool

Contact

For questions, issues, or collaboration related to this dataset, please open an issue or contact the repository maintainer.