Research on Chain-of-Thought Prompting Enhances Code Deobfuscation Using Large Language Models

A study explores Chain-of-Thought prompting to improve code deobfuscation, a process that restores readable program versions while maintaining original behavior. This method guides large language models through step-by-step reasoning for code analysis, focusing on control flow obfuscation techniques like Control Flow Flattening and Opaque Predicates. Evaluation of five state-of-the-art models shows significant quality improvements over simple prompting. The research validates the approach on diverse C benchmarks, measuring structural recovery of control flow graphs and semantic preservation. Code deobfuscation typically demands extensive manual effort with complex tools, but this alternative reduces that burden. Results are reported using structural metrics for control flow, highlighting practical applications in software analysis.