Reproduce “Unmasking the Shadows: Pinpoint the Implementations of Anti-Dynamic Analysis Techniques in Malware Using 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

Haizhou Wang, Nanqing Luo, Xusheng Li, Peng LIu

Researchers from Pennsylvania State University and Vector 35 Inc. developed an LLM-based static analysis workflow to pinpoint Anti-Dynamic Analysis Technique (TADA) implementations in malware binaries at the basic block level. The method achieved an 87.80% detection rate for known TADAs and substantially reduced the code search space for human reverse engineers, allowing for more efficient identification of debugger breakpoints.

Key insights

  • Large Language Models can effectively bridge the semantic gap between low-level binary code and natural language for security analysis when provided with carefully engineered, contextual features.
  • The processing of strings, including deobfuscated ones, is critical for robust TADA detection, as many evasion techniques rely on environmental artifact checks via string comparisons, leveraging LLMs' natural language understanding capabilities.
  • Pinpointing TADA implementations at the basic block level offers a highly practical and actionable output for reverse engineers, directly enabling more efficient placement of breakpoints to bypass evasive code.

Summary by alphaXiv.

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