Trustworthy AI for Security Decision-Making in ICSs: Towards Compliance with the EU AI Act

Abstract

In Industrial Control Systems (ICSs), security is not a fancy add-on feature but a core requirement for functionality, safe operation, and organization business. To keep up with the ever-growing threat landscape and numerous security standards and regulations, leveraging Artificial Intelligence (AI) capabilities, especially Large Language Models (LLMs), seems essential. However, AI systems built using LLMs can introduce new challenges such as data leakage, generating bias and misinformation, and even new security threats. In this paper, we analyze an AI system that leverages LLMs and knowledge graphs to support security decisions in the design of ICSs, ensuring compliance with the IEC 62443-3-3 standard while aligning with the European AI Act to guarantee trustworthiness and meet legal requirements for use in Europe.