LLM assisted reverse engineering
Points and comments are a snapshot, not live.
An experienced reverse engineer details how LLMs dramatically accelerate hacking hardware and software.
The author, with two decades of reverse engineering experience, now uses LLMs (primarily DeepSeek v4.1 Flash with Ghidra and radare2) to greatly speed up the process. Their workflow involves having the agent document findings to markdown files, analyze binary structure, and use hypothesis-driven debugging when stuck. The author has achieved successes including reverse engineering the Alphasmart NEO, mapping communication protocols for Sony cameras and biofeedback devices, developing custom firmwares for synthesizers (Arturia MiniFreak, Teenage Engineering OP-Z, Akai MPC), and cracking abandoned software. They use Python with UV tooling for prototyping and Rust for final implementations.
What commenters are saying
Commenters broadly agreed that LLMs are powerful for reverse engineering abandoned or proprietary software and hardware. Several shared their own successful projects, including reverse engineering an audio plugin and optimizing Linux on an old Intel Mac. A split emerged between those who find IDA Pro superior to Ghidra for LLM-assisted RE, and others who prefer Ghidra's emulation capabilities. A significant sub-thread discussed circumventing LLM guardrails against RE, with some using DeepSeek to bypass restrictions, and others noting Claude's recent increased refusal behavior. One commenter warned about legal agreements like Apple's developer terms, while others dismissed such concerns.