To thwart reverse engineering by agents, dev proposes blinding them

Timo Zhang, a software developer and student, has developed a project called Blind LLM Agents (BLA) to hinder AI agents from reverse-engineering commercial software binaries. LLMs can now analyze code and recreate its functionality, making BLA a potential safeguard for developers concerned about unauthorized copying of their work.

Zhang explained that LLMs can reverse-engineer shipped packages, CLIs, and WebAssembly in minutes, rendering classic obfuscation ineffective against AI-driven analysis. BLA applies various techniques to disrupt reverse-engineering efforts, including renaming, string encryption, build polymorphism, decoys, self-seal integrity checks, and a coverage matrix.

While BLA doesn't prevent committed reverse engineering, it introduces additional challenges for LLM-assisted analysis, potentially increasing the cost of breaking the protection. Zhang's project has faced criticism for potentially limiting the availability of publicly accessible binaries, but BLA is designed specifically to protect shipping code from reverse engineering and LLM-aided analysis.

Zhang has compared BLA to other projects aiming to hinder AI interference, such as Poison Fountain and Anubis, but emphasizes that BLA is focused on safeguarding hardened code against reverse engineering.