LLMs Assess Air Safety at Non-Towered Airports

A recent study released on arXiv suggests the application of large language models (LLMs) and vision-language models (VLMs) to enhance post-flight safety evaluations at non-towered airports, which do not have air traffic control towers. These facilities depend on the Common Traffic Advisory Frequency (CTAF) for pilots to announce their positions, resulting in numerous near mid-air collisions. The proposed framework examines transcribed CTAF communications, METAR weather reports, ADS-B flight paths, and VFR sectional charts. A preliminary investigation at Half Moon Bay Airport features both a qualitative case study and a quantitative analysis using synthetic data. Testing with Gemini 2.5 Pro revealed effective identification of safety concerns, aiming to boost safety assessment processes at thousands of non-towered airports globally.