Deobfuscating Embedded Malware Using Probable-Plaintext Attacks

Malware embedded in documents is regularly used as part of targeted attacks. To hinder a detection by anti-virus scanners, the embedded code is usually obfuscated, often with simple Vigenere ciphers based on XOR, ADD and additional ROL instructions. While for short keys these ciphers can be easily cracked, breaking obfuscations with longer keys requires manually reverse engineering the code or dynamically analyzing the documents in a sandbox. In this paper, we present Kandi, a method capable of efficiently decrypting embedded malware obfuscated using Vigenere ciphers. To this end, our method performs a probable-plaintext attack from classic cryptography using strings likely contained in malware binaries, such as header signatures, library names and code fragments. We demonstrate the efficacy of this approach in different experiments. In a controlled setting, Kandi breaks obfuscations using XOR, ADD and ROL instructions with keys up to 13 bytes in less than a second per file. On a collection of real-world malware in Word, Powerpoint and RTF files, Kandi is able to expose obfuscated malware from every fourth document without involved parsing.