Modeling and Engineering Cyber–Safety Causality in Autonomous Maritime Navigation

Abstract

The increasing deployment of Unmanned Surface Vessels (USVs) and Maritime Autonomous Surface Ships (MASS) requires tighter integration of safety and cyber-security analysis. Current approaches often treat these concerns separately, overlooking the causal pathways through which cyberattacks can degrade safety-critical functions. This paper addresses this gap using the Structural Causal World Model (SCWM), a layered causal modelling framework with explicit uncertainty semantics and safety-metric bindings. SCWM represents system attributes as nodes in a semantically typed directed acyclic causal structure and enables the extraction of attack-induced propagation subgraphs linking cyber-security violations to collision risk indicators. We present the MASS-SCWM framework and demonstrate its application to compound attacks targeting vessel communications and vision-based perception systems, including loss of connectivity with the Remote Operations Centre and misclassification or range estimation errors in object detection. These attacks are modelled as attribute-level perturbations whose effects propagate through belief dynamics and decision layers to safety metrics such as Closest Point of Approach (CPA) and Time to CPA (TCPA). The framework enables structured and quantitatively comparable analysis of compound attack scenarios and protection mechanisms. By formally binding cyber-security assumptions to safety guarantees within a unified causal structure, MASS-SCWM provides a principled foundation for integrated safety–security assurance aligned with emerging regulatory expectations for MASS systems.