From Risk Assessments to Knowledge Statements: Epistemic Uncertainty in LLM-Based Risk Management for Critical Infrastructures
From Risk Assessments to Knowledge Statements: Epistemic Uncertainty in LLM-Based Risk Management for Critical Infrastructures

Summary/Abstract: The growing use of large language models (LLMs) in risk management is often justified by improved efficiency and better handling of unstructured data. However, the epistemic quality of the resulting risk statements remains poorly explored. This article investigates epistemic uncertainty in LLM-based risk management from a conceptual perspective, drawing on risk theory and epistemological concepts. LLM-generated risk statements are analyzed as knowledge statements and assessed in terms of their epistemic viability and knowledge strength through a theory-driven literature review. The analysis shows that LLMs do not reduce epistemic uncertainty; instead, they alter its visibility and assessability. Fluent risk statements may hide implicit assumptions and context-specific knowledge gaps, leading to an overestimation of their usefulness for decision-making. This is especially concerning in safety-critical areas, like critical infrastructure, where risk management decisions are closely tied to national security. Without an explicit epistemic context, risk statements produced by LLMs do not provide enough foundation for risk-based decisions.

• Details
• Contents