LLM-Assisted Dynamic Threat Analysis for Attacker-Reachable Software Weaknesses in Autonomous Vehicles

Md Wasiul Haque, Sagar Dasgupta, Mizanur Rahman, Md Rayhanur Rahman|August 13, 2026arXiv

Key Takeaway

LLM-generated test harnesses for safety-critical autonomous vehicle code fail primarily due to build system complexity, not reasoning ability—suggesting that tooling and integration matter more than model capability for dynamic vulnerability analysis.

Summary

This paper investigates whether LLMs can automatically generate executable test artifacts to confirm exploitable weaknesses in Autoware, an autonomous vehicle software stack.

safety evaluation

Key Terms

static-analysis dynamic-analysis fuzzing compiler-in-the-loop threat-analysis