Detecting and Categorizing Android Malware with Graph ...

Peng Xu1, Claudia Eckert1, Apostolis Zarras2

{Peng,eckert}sec.in.tum.de

[email protected] Technical University of Munich2 Delft University of Technology

Detecting and Categorizing Android Malware with Graph Neural Networks

2Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

https://securelist.com/mobile-malware-evolution-2020/101029/

3Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

G DATA Mobile Malware Report 2019: New high for malicious Android apps

4Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

1. permission-based Android Malware Detection systems (DREBIN, FM)2. API-call-based Android Malware Detection systems (DroidNative)

https://developer.android.com/reference/android/Manifest.permissionOpCode-Level Function Call Graph Based Android Malware Classification Using Deep Learning

String, Opcode(word)

5Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

String ObfuscationString(permission), API Call(word)

• Class Encryption

• String Encryption

• Reflection• replace each invoke instruction with specific

bytecode

• Trivial Obfuscation• Only affects string, not bytecode

• Trivial + String Encryption

• Trivial + StringEnc + Reflection

Obfuscation

• Trivial + StringEnc + Reflection + ClassEnc

6Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

https://www.microsoft.com/security/

7Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

DroidOL: Android malware detection based on online machine learning

8Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

Adagio: Structural Detection of Android Malware using Embedded Call-GraphMANIS: evading malware detection system on graph structure

9Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

Neural Network-based Graph Embedding for Cross-Platform Binary Code Similarity Detection

10Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Motivation

11Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Overview

APK fileFunction call graph

with opcodeembedding

Function call graphwith functionembedding

GraphEmbedding

Function callgraph

2-layer MLPMalware detection

2-layer MLPMalware classification

Benign

Malware

Plankton

FakeInstaller

DroidKungFu

...

MLP

12Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Function Call Graph

Androguard to get Function call graph(e.g, Adagio, MANIS)

• Instruction: Opcode + Operands• Why only consider Opcode?

• Other works: Address, Register are replaced by specificsymbols

• Move Instruction: move-wide vA, vB[04 12x], move-wide/from16 vAA, vBBBB[05 22x]

• Invoke Instruction: invoke-super, invoke-direct, invoke-static, and invoke-interface

• Word Embedding

13Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Opcode Embedding

• Weighted Mean Function Embedding

• SIF-Invoked Function Embedding • SIF: A simple but tough-to-beat baseline for sentence embeddings.

14Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

Function Embedding

15Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Graph Embedding

16Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security, Technical University of Munich

MLP Classifier

• Malware Classification:

• Malware Categorization:

17Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation

18Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Various learning rate

19Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Various training Epoch

20Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Various training Epoch

21Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Various n-hop neighbors

22Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Obfuscated Application

23Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Obfuscated Application

24Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Evaluation – Categorization/Family Classification

25Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Question?Thanks!

26Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Backup

27Peng Xu, Claudia Eckert, Apsotlis Zarras | IT Security | Technical University of Munich

Backup – Structure2vec