As digitalized systems are the backbone of smart grids, their vulnerability to cyberattacks grows—evident in high-profile incidents like Ukraine’s power grid attacks and the 2022 Viasat breach. While AI-driven cybersecurity is gaining traction, the role of Large Language Models (LLMs) in protecting critical infrastructure remains largely unexplored. By systematically quantifying LLMs capabilities in offensive and defensive critical infrastructure cybersecurity, this research delivers insights into their strengths, limitations, and real-world applicability. The findings pave the way for more intelligent, automated, and resilient cybersecurity solutions in the ever-evolving threat landscape.