This article explores the prospects for integrating machine learning (ML) methods into the architecture of digital twins and intelligent operator support systems (IOSS) for nuclear power plant (NPP) units in the context of the transition to Industry 5.0. The relevance of the study is determined by the need to improve the adaptability and predictive capabilities of existing NPP control systems, as well as to strengthen the human-centric design of operator interfaces. The paper analyzes modern approaches using deep learning algorithms, neural networks, and natural language processing methods to solve diagnostic problems, predict anomalies, and generate recommendations for operators. Architectural solutions for integrating ML models with existing components of digital twins of NPP units are considered, including issues of ensuring the interpretability of artificial intelligence (AI) solutions. Particular attention is paid to the problem of operator trust in the recommendations of intelligent systems and methods for its formation through the mechanisms of explainable AI. A conceptual model of an adaptive IOSS is proposed that takes into account the operator’s individual cognitive characteristics and provides a personalized presentation of information. The study’s results demonstrate that integrating ML methods with Industry 5.0 principles enables a qualitatively new level of interaction between humans and intelligent systems in critical infrastructures.