Ethical and governance challenges of large language models in power and energy systems: a review and framework for responsible AI deployment

Abstract

Large Language Models (LLMs) are increasingly used for forecasting, monitoring, diagnostics, predictive maintenance, and decision support in power and energy systems, yet their deployment in safety-critical infrastructures raises concerns regarding hallucination, bias, explainability, cybersecurity, accountability, and human oversight. This structured scoping and critical review synthesizes a 126-source evidence corpus and compares emerging LLM paradigms, including domain-adapted models, retrieval-augmented generation, multimodal and reasoning models, agentic systems, and MCP-enabled architectures. Particular attention is given to power-system-specific requirements such as SCADA/EMS/DMS integration, topology awareness, state estimation, power-flow and voltage constraints, N-1 security, protection coordination, and operator authority. Quantitative publication trends and comparative evidence analysis are used to identify methodological and deployment gaps in the literature. Building on these findings, a multi-layer governance framework is proposed that integrates data and model governance, deterministic engineering verification, cybersecurity safeguards, human authorization, auditability, and regulatory oversight. A distribution-network fault-diagnosis and service-restoration workflow, together with a staged validation protocol, illustrates practical implementation. The review concludes with measurable research priorities for trustworthy LLM-enabled energy systems. © 2026 The Authors.