LLM assisted reverse engineering
A two-decade hardware and software hacker describes using LLMs to accelerate reverse engineering, pointing an agent at binaries and driving it through Ghidra's MCP bridge or radare2. The agent documents findings to Markdown files, maps structure, and when stuck debugs, adds logging or probes, and follows a hypothesis-driven loop of testable experiments logged to CSV. He credits the approach with cracking an Alphasmart NEO, camera and biofeedback protocols, synth firmwares, and abandonware licenses. Commenters largely corroborate, citing quick wins on CCTV plugins, Android apps, and audio plugins, though several flag the agent's guessing loops as the main frustration and one argues the process, not the result, is what builds skill.
Original article ↗·HN comments ↗
slowchronicle.com · 301 points · 77 comments · ioma8 · Nov 18, 2026