Reproduce “Disassembling Obfuscated Executables with LLM” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Huanyao Rong, Yue Duan, Hang Zhang, XiaoFeng Wang, Hongbo Chen, Shengchen Duan, Shen Wang

DISASLLM introduces a novel approach to improve binary disassembly accuracy for obfuscated executables by integrating an LLM-based validity classifier into a multi-stage analysis workflow. The system achieved approximately 40% improvement in detecting the first instruction after a junk byte region compared to state-of-the-art disassemblers and demonstrated strong resilience against various obfuscation techniques.

Key insights

  • An LLM-based classifier can effectively determine the semantic validity of assembly instructions within their context, surpassing traditional methods that struggle with complex obfuscation.
  • Integrating an LLM validity classifier into a multi-stage disassembly workflow, including initial parsing, validity checking, and error correction (infilling), enhances accuracy.
  • Contextual information from surrounding instructions significantly improves the LLM classifier's ability to distinguish valid code from obfuscated elements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.