Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Applying Large Language Models to Power Systems: Potential Security Threats

Jiaqi Ruan, Gaoqi Liang, Huan Zhao, Guolong Liu, Xianzhuo Sun, Jing Qiu, Zhao Xu, Fushuan Wen, Zhao Yang Dong

TL;DR

This paper investigates the security implications of applying large language models (LLMs) to cyber-physical power systems (CPPS), recognizing potential gains in data analytics, forecasting, and operator interaction alongside emerging security risks. It provides a threat taxonomy across privacy invasion, performance integrity, semantic divergence, and denial-of-service, illustrating how open CPPS interfaces could be exploited. The authors propose a multi-layered mitigation framework including secure LLM architectures, anomaly detection, data sanitization, access controls, and human-in-the-loop governance, aligned with cybersecurity standards and regulatory requirements. The work aims to guide researchers and practitioners toward secure, reliable deployment of LLM-enabled decision support in future power systems.

Abstract

Applying large language models (LLMs) to modern power systems presents a promising avenue for enhancing decision-making and operational efficiency. However, this action may also incur potential security threats, which have not been fully recognized so far. To this end, this article analyzes potential threats incurred by applying LLMs to power systems, emphasizing the need for urgent research and development of countermeasures.

Applying Large Language Models to Power Systems: Potential Security Threats

TL;DR

This paper investigates the security implications of applying large language models (LLMs) to cyber-physical power systems (CPPS), recognizing potential gains in data analytics, forecasting, and operator interaction alongside emerging security risks. It provides a threat taxonomy across privacy invasion, performance integrity, semantic divergence, and denial-of-service, illustrating how open CPPS interfaces could be exploited. The authors propose a multi-layered mitigation framework including secure LLM architectures, anomaly detection, data sanitization, access controls, and human-in-the-loop governance, aligned with cybersecurity standards and regulatory requirements. The work aims to guide researchers and practitioners toward secure, reliable deployment of LLM-enabled decision support in future power systems.

Abstract

Applying large language models (LLMs) to modern power systems presents a promising avenue for enhancing decision-making and operational efficiency. However, this action may also incur potential security threats, which have not been fully recognized so far. To this end, this article analyzes potential threats incurred by applying LLMs to power systems, emphasizing the need for urgent research and development of countermeasures.
Paper Structure (9 sections, 1 figure)

This paper contains 9 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Potential Threats of Large Language Models in Power Systems
  3. Large Language Models in Power Systems
  4. Potential Threats Incurred by Applying Large Language Models
  5. Privacy invasion through large language models
  6. Deteriorated performance in large language models
  7. Threats from semantic divergence
  8. Denial of service for large language models
  9. Conclusion and Suggestions

Figures (1)

  • Figure 1: Illustration of applying the large language model to the cyber-physical power system.