We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them

The article discusses experiments in using AI and the Ghidra reverse engineering tool to detect hidden backdoors in binary executables. While the AI agents were able to find some obvious backdoors, the approach is not yet ready for production use, as even the best model detected malicious code in small/mid-size binaries only 49% of the time and had a high false positive rate. The article also provides background on recent supply chain attacks and security issues with firmware, and explores the challenges of binary analysis compared to working with human-readable source code.