CyberLLM AI system enhances automotive software vulnerability detection

Researchers from the Technical University of Munich and industrial partners have developed CyberLLM, an artificial intelligence system designed to enhance automotive cybersecurity. As vehicles transition into software-defined vehicles (SDVs), the digital attack surface expands to include source code, onboard networks, and over-the-air (OTA) updates.

CyberLLM utilizes multi-agent large language models (LLMs) to autonomously identify vulnerabilities in automotive software and plan potential responses. In testing involving nine electronic control units (ECUs) written in C, C++, and Rust, the system demonstrated significant improvements in detection capabilities. While traditional deterministic inspection methods identified 34% of 47 embedded vulnerabilities, the addition of LLM-based analysis increased the detection rate to approximately 70%. The research indicates that the system produced no false positives when testing pure control modules.