An automated Bayesian network framework for maritime cybersecurity risk assessment powered by large language models

Zhao, Y, Li, H ORCID: 0000-0002-4293-4763, Thanh Nguyen, T, Matthews, C ORCID: 0000-0002-4126-6484 and Yang, Z ORCID: 0000-0003-1385-493X (2026) An automated Bayesian network framework for maritime cybersecurity risk assessment powered by large language models. Safety Science, 204. pp. 1-17. ISSN 0925-7535

Preview

Text An automated Bayesian network framework for maritime cybersecurity risk assessment powered by large language models.pdf - Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

Abstract

The accelerating digitalisation of the maritime sector has increased its operational efficiency while simultaneously heightening its vulnerability to cyber-induced crises and emergency situations. Cyber incidents such as ransomware attacks, Global Positioning System (GPS) jamming, and phishing can trigger cascading system-level disruptions, undermining maritime operations, safety, and the resilience of critical transport infrastructures. To address this challenge, this study proposes an automated and scalable Bayesian Network (BN) framework powered by Large Language Models (LLMs) to support decision-relevant maritime cybersecurity risk analysis and assessment. A LLM-based text-mining pipeline is developed to reconstruct a comprehensive maritime cyber incident dataset (2001–2025) from unstructured reports. The proposed pipeline enables systematic identification and classification of key Risk Influential Factors (RIFs), which are subsequently validated through accuracy and consistency evaluations to ensure reliability and auditability. These RIFs are then modelled as variables within a BN model, facilitating probabilistic inference, characterization of interdependencies, and explicit analysis of risk propagation pathways associated with cyber-induced operational disruptions. Comparative experiments demonstrate that the proposed LLM-BN framework outperforms conventional BN models in terms of data consistency, inference accuracy, and structural interpretability. Beyond methodological contributions, the framework provides actionable insights to support emergency preparedness, risk prioritisation, and resilience enhancement for maritime operators, infrastructure managers, and regulators. By integrating structured knowledge extracted from textual sources with probabilistic risk modelling, this study contributes a decision-support framework for understanding, assessing, and managing cyber risks in complex socio-technical maritime systems, with broader applicability to other critical infrastructure sectors.

Item Type:	Article
Uncontrolled Keywords:	Maritime cybersecurity; Large language model; Bayesian network; Risk influential factors; Probabilistic inference; 40 Engineering; Generic health relevance; 09 Engineering; 11 Medical and Health Sciences; 17 Psychology and Cognitive Sciences; Human Factors; 40 Engineering; 42 Health sciences; 52 Psychology
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering
Divisions:	Engineering and Built Environment
Publisher:	Elsevier
Date of acceptance:	16 July 2026
Date of first compliant Open Access:	10 September 2026
Date Deposited:	10 Sep 2026 09:58
Last Modified:	10 Sep 2026 09:58
DOI or ID number:	10.1016/j.ssci.2026.107384
URI:	https://researchonline.ljmu.ac.uk/id/eprint/29373

View Item