|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- cybersecurity |
|
|
- malware |
|
|
- disassembly |
|
|
pretty_name: EMBER2024-capa |
|
|
size_categories: |
|
|
- 10M<n<100M |
|
|
--- |
|
|
|
|
|
# EMBER2024-capa Dataset |
|
|
|
|
|
[Capa](https://github.com/mandiant/capa) is a malware analysis tool that identifies a file's capabilities: |
|
|
|
|
|
``` |
|
|
+--------------------------------------+-----------------------------------------+ |
|
|
| CAPABILITY | NAMESPACE | |
|
|
|--------------------------------------|-----------------------------------------| |
|
|
| reference anti-VM strings | anti-analysis/anti-vm/vm-detection | |
|
|
| encode data using XOR (2 matches) | data-manipulation/encoding/xor | |
|
|
| contain an embedded PE file | executable/subfile/pe | |
|
|
| run as a service | executable/pe | |
|
|
| create service | host-interaction/service/create | |
|
|
| persist via Windows service | persistence/service | |
|
|
+--------------------------------------+-----------------------------------------+ |
|
|
``` |
|
|
|
|
|
We ran Capa on all of the malicious Win32 and Win64 files in the [EMBER2024 dataset](https://huggingface.co/datasets/joyce8/EMBER2024). In many cases, Capa identifies the function(s) that the capability is present in. We identified 583 distinct capabilities in 16,356,790 functions (10,756,829 from Win32 files, 5,599,961 from Win64 files). The dataset includes a JSON object for each function. For example: |
|
|
|
|
|
``` |
|
|
{ |
|
|
"sha256": "0046238d6069959f2190d7d1362c3748959e8befa8b3d09d5329f2cdc04b551e", |
|
|
"func_addr": "0x46f40c", |
|
|
"capa": ["Log keystrokes via polling"], |
|
|
"bytes": "558bec53568bf133db399e68050000570f84a4000000817d08000100008b7d0c755e6683ff0d75466a11ff1588644e006685c07d21389e6f05000075198b86480400003bc3740b53536a105650e8ee67ffff33c0eb67389e6e05000075428b86480400003bc374ea53536a0febdd8b86480400003bc37428ff7510576a14eb19817d080101000075178b86480400003bc3740dff7510576a135650e8a067ffffff75148b8668050000ff75108b480c8b0157ff7508ff500ceb0333c0405f5e5b5dc21000", |
|
|
"disasm": ["push ebp", "mov ebp, esp", "push ebx", "push esi", "mov esi, ecx", "xor ebx, ebx", "cmp dword [esi + 0x568], ebx", "push edi", "je 0x46f4c6", "cmp dword [ebp + 8], 0x100", "mov edi, dword [ebp + 0xc]", "jne 0x46f48c", "cmp di, 0xd", "jne 0x46f47a", "push 0x11", "call dword [sym.imp.USER32.dll_GetKeyState]", "test ax, ax", "jge 0x46f462", "cmp byte [esi + 0x56f], bl", "jne 0x46f462", "mov eax, dword [esi + 0x448]", "cmp eax, ebx", "je 0x46f45e", "push ebx", "push ebx", "push 0x10", "push esi", "push eax", "call fcn.00465c4c", "xor eax, eax", "jmp 0x46f4c9", ..."ret 0x10"]} |
|
|
``` |
|
|
|
|
|
The dataset is not de-duplicated. Identical functions may appear in different files (or even in the same file!). Capabilities do not necessarily indicate malicious behaviors, but they are generally useful to be aware of when analyzing malware. Some capabilities are far more common than others. The top 10 capabilities in EMBER2024-capa are: |
|
|
|
|
|
| Capability | Count | |
|
|
-------------|------------------ |
|
|
| Encode data using xor | 1,403,283 | |
|
|
| Read file on windows | 1,117,738 | |
|
|
| Resolve function by parsing pe exports | 1,069,762 | |
|
|
| Get common file path | 894,799 | |
|
|
| Write file on windows | 863,598 | |
|
|
| Query or enumerate registry value | 744,832 | |
|
|
| Check if file exists | 668,035 | |
|
|
| Delete file | 659,308 | |
|
|
| Get file attributes | 580,237 | |
|
|
| Create process on windows | 522,145 | |
|
|
|
|
|
For more Capa data (including Mitre ATT&CK TTPs, Malware Behavior Catalog Objectives and Behaviors, and more capabilities, see the full [EMBER2024 dataset](https://github.com/futurecomputing4ai/ember2024). If you use this dataset or the full EMBER2024 dataset in your research, please cite: |
|
|
|
|
|
``` |
|
|
@inproceedings{joyce2025ember2024, |
|
|
title={EMBER2024-A Benchmark Dataset for Holistic Evaluation of Malware Classifiers}, |
|
|
author={Joyce, Robert J and Miller, Gideon and Roth, Phil and Zak, Richard and Zaresky-Williams, Elliott and Anderson, Hyrum and Raff, Edward and Holt, James}, |
|
|
booktitle={Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2}, |
|
|
pages={5516--5526}, |
|
|
year={2025} |
|
|
} |
|
|
``` |
|
|
|
