Evaluating Open-Weight LLMs for Generating Structured Threat Information for Autonomous Vehicle Vulnerabilities

Md Erfan, Ahmed Ryan, Md Kamal Hossain Chowdhury, Md Rayhanur Rahman|July 17, 2026arXiv

Key Takeaway

Open-weight LLMs can automate threat intelligence extraction from vulnerability databases, achieving 94% accuracy on basic threat objects but only 63% on relationships—useful for scaling security analysis but requiring human review for complex threat patterns.

Summary

This paper evaluates open-weight LLMs on converting vehicle vulnerability descriptions into STIX (Structured Threat Information Expression) format—a standardized way to represent security threats.

safety evaluation applications

Key Terms

cve stix cwe mitre-att-ck