Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Risks of Practicing Large Language Models in Smart Grid: Threat Modeling and Validation

Jiangnan Li, Yingyuan Yang, Jinyuan Sun

TL;DR

This work addresses the security risks of deploying large language models in smart grids by formalizing two threat models: bad data injection and domain knowledge extraction. It presents a threat-analysis framework grounded in the prompt-based operation of LLMs and validates the models across multiple popular LLMs (e.g., GPT-3.5, GPT-4, Gemini 2.0, LLaMA-3, DeepSeek-V3) using real grid data. Validation shows attackers can inject misleading inputs to degrade classification performance and can extract domain knowledge under crafted prompts, highlighting practical vulnerabilities in smart-grid contexts. The authors open-source data, code, and results to spur further research and defense development, signaling a need for robust prompt governance and secure deployment practices in critical infrastructure. The work provides a concrete, actionable baseline for evaluating and mitigating LLM-related risks in smart grids.

Abstract

Large language models (LLMs) represent significant breakthroughs in artificial intelligence and hold potential for applications within smart grids. However, as demonstrated in previous literature, AI technologies are susceptible to various types of attacks. It is crucial to investigate and evaluate the risks associated with LLMs before deploying them in critical infrastructure like smart grids. In this paper, we systematically evaluated the risks of LLMs and identified two major types of attacks relevant to potential smart grid LLM applications, presenting the corresponding threat models. We validated these attacks using popular LLMs and real smart grid data. Our validation demonstrates that attackers are capable of injecting bad data and retrieving domain knowledge from LLMs employed in different smart grid applications.

Risks of Practicing Large Language Models in Smart Grid: Threat Modeling and Validation

TL;DR

This work addresses the security risks of deploying large language models in smart grids by formalizing two threat models: bad data injection and domain knowledge extraction. It presents a threat-analysis framework grounded in the prompt-based operation of LLMs and validates the models across multiple popular LLMs (e.g., GPT-3.5, GPT-4, Gemini 2.0, LLaMA-3, DeepSeek-V3) using real grid data. Validation shows attackers can inject misleading inputs to degrade classification performance and can extract domain knowledge under crafted prompts, highlighting practical vulnerabilities in smart-grid contexts. The authors open-source data, code, and results to spur further research and defense development, signaling a need for robust prompt governance and secure deployment practices in critical infrastructure. The work provides a concrete, actionable baseline for evaluating and mitigating LLM-related risks in smart grids.

Abstract

Large language models (LLMs) represent significant breakthroughs in artificial intelligence and hold potential for applications within smart grids. However, as demonstrated in previous literature, AI technologies are susceptible to various types of attacks. It is crucial to investigate and evaluate the risks associated with LLMs before deploying them in critical infrastructure like smart grids. In this paper, we systematically evaluated the risks of LLMs and identified two major types of attacks relevant to potential smart grid LLM applications, presenting the corresponding threat models. We validated these attacks using popular LLMs and real smart grid data. Our validation demonstrates that attackers are capable of injecting bad data and retrieving domain knowledge from LLMs employed in different smart grid applications.
Paper Structure (18 sections, 6 figures, 2 tables)

This paper contains 18 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Machine Learning and NLP in Power Grids
  4. AI Security in Smart Grid
  5. Threat Analysis and Modeling
  6. Background: Prompt and LLM
  7. Threat Analysis: Scope
  8. Threat Models
  9. Bad Data Injection
  10. Domain Knowledge Extraction
  11. Validation
  12. Bad Data Injection
  13. Application Introduction
  14. Attack Simulation
  15. Domain Knowledge Extraction
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Illustration of cyberattacks to LLMs used in smart grid. The figure shows two cases: (1) an outsider attacker injects false information into LLMs and (2) an internal attacker obtains unauthorized information from LLMs.
  • Figure 2: LLM Application Workflow
  • Figure 3: Message Samples
  • Figure 4: Bad Data Injection Prompts Presentation
  • Figure 5: System Prompt (Domain Knowledge) for LLM-based Virtual Assistant
  • ...and 1 more figures