USING LLMS AS A REVERSE ENGINEERING SIDEKICK

Summary from the TLDR InfoSec newsletter

This post demonstrates how to use LLMs with Model Context Protocol (MCP) servers integrated into IDA Pro and Ghidra to accelerate malware reverse engineering through automated function analysis, code documentation, and variable renaming. Testing on IcedID malware compared cloud-based Claude Sonnet 3.7 ($2.91-$13.24 per analysis, 18-11 minutes) against local Ollama/Devstral models ($0 cost, 22-46 minutes), with cloud models providing more thorough analysis but at a higher expense. Key limitations include token costs escalating with file complexity, context window truncation causing local models to forget instructions mid-analysis, and security risks from prompt injection, tool poisoning, and potential data exfiltration when using untrusted MCP servers.