Comparative Analysis of Human and Independent Large Language Model (LLM) Perspectives on the Top Ten Cybersecurity Issues in Aviation

Abstract

This study compares cybersecurity threat prioritizations produced by aviation subject-matter experts and ten independently developed large language models (LLMs). Using Borda ranking methods and Kendall’s W to evaluate agreement, we analyze aligned themes and divergences in aviation-specific threat emphasis. SMEs prioritized risks at integration boundaries, safety-critical navigation interference, and supply-chain and cloud-to-aircraft trust paths. LLMs successfully identified broad risk categories but emphasized generic IT attack surfaces more heavily than domain-specific vectors. Findings suggest LLMs provide value for horizon scanning and taxonomy scaffolding but require aviation context to support operational prioritization. Limitations include a small SME sample (n=4) and rapidly evolving AI model capabilities. Future work should expand stakeholder representation across operators, ANSPs, airport authorities, and regulators.