From Incomplete Architecture to Quantified Risk: Multimodal LLM-Driven Security Assessment for Cyber-Physical Systems DOI: XXXXXXX.XXXXXXX CCS: Security and privacy Malware and its mitigation CCS: Computing methodologies Knowledge representation and reasoning CCS: Computer systems organization Embedded and cyber-physical systems CCS: Security and privacy Vulnerability management

Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar · 2026

Read From Incomplete Architecture to Quantified Risk: Multimodal LLM-Driven Security Assessment for Cyber-Physical Systems DOI: XXXXXXX.XXXXXXX CCS: Security and privacy Malware and its mitigation CCS: Computing methodologies Knowledge representation and reasoning CCS: Computer systems organization Embedded and cyber-physical systems CCS: Security and privacy Vulnerability management by Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-05674.txt.

Introduction

Huang, Poskitt, and Shar rate cyber-physical safety from patchy architecture papers with ASTRAL, a multimodal language-model tool. Prompt chaining and design reasoning rebuild system views from scattered sources, enabling adaptive threat spotting and counted risk scores. Ablation tests plus review by 14 experts rate the approach useful and sound.

Catalog record

CREATORShaofei Huang, Christopher M. Poskitt, Lwin Khin Shar
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE87151
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.05674

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

From Incomplete Architecture to Quantified Risk: Multimodal LLM-Driven Security Assessment for Cyber-Physical Systems DOI: XXXXXXX.XXXXXXX CCS: Security and privacy Malware and its mitigation CCS: Computing methodologies Knowledge representation and reasoning CCS: Computer systems organization Embedded and cyber-physical systems CCS: Security and privacy Vulnerability management Shaofei Huang, Christopher M. Poskitt, Lwin Khin Shar Abstract . Cyber-physical systems often contend with incomplete architectural documentation or outdated information resulting from legacy technologies, knowledge management gaps, and the complexity of integrating diverse subsystems over extended operational lifecycles. This architectural incompleteness impedes reliable security assessment, as inaccurate or missing architectural knowledge limits the identification of system dependencies, attack surfaces, and risk propagation pathways. To address this foundational challenge, this paper introduces ASTRAL (Architecture-Centric Security Threat Risk Assessment using LLMs), an architecture-centric security assessment technique implemented in a prototype tool powered by multimodal LLMs. The proposed approach assists practitioners in reconstructing and analysing CPS architectures when documentation is fragmented or absent. By leveraging prompt chaining, few-shot learning, and architectural reasoning, ASTRAL extracts and synthesises system representations from disparate data sources. By integrating LLM reasoning with architectural modelling, our approach supports adaptive threat identification and quantitative risk estimation for cyber-physical systems. We evaluated the approach through an ablation study across multiple CPS case studies and an expert evaluation involving 14 experienced cybersecurity practitioners. Practitioner feedback suggests that ASTRAL is useful and reliable for supporting architecture-centric security assessment. Overall, the results indicate that the approach can support more informed cyber risk management decisions. 1. Introduction In cyber-physical systems (CPSs), software architecture forms the structural foundation upon which system functionality, performance, and security depend. Architectural models play a pivotal role in security assessment by providing a coherent representation of components, interfaces, and data flows within complex, multi-layered environments ( 27 ) . Decisions made at the architectural level, such as component decomposition, communication pathways, and trust boundary definition, directly shape the system’s attack surface and influence how vulnerabilities emerge and propagate. These structural choices determine not only where defensive controls can be placed, but also how effectively threats can be isolated, contained, or mitigated. Architectural patterns that emphasise isolation, redundancy, and secure communication therefore enhance system resilience and reduce the likelihood of cascading failures. Effective security assessment is thus inseparable from architectural analysis, as the same structural decisions that govern system behaviour and performance also define the conditions under which security controls can be systematically integrated and enforced throughout the system life-cycle. In real-world settings, practitioners frequently encounter significant impediments due to architectural incompleteness , where documentation is fragmented, outdated, or entirely absent ( 18 ) . This gap in architectural knowledge disrupts security assessment workflows, particularly within frameworks that presuppose the existence of exhaustive architectural documentation. CPS environments often contend with these gaps due to extended operational life-cycles, persistence of legacy technologies, and complexities of multi-vendor subsystem integration. These factors result in missing details about components, versions, and interconnections ( 53 ; 49 ) . Furthermore, threat models developed during the initial design phase rapidly become obsolete as the physical and digital architectures evolve through iterative updates ( 21 ; 26 ) , creating a discontinuity that complicates operational security assessment. Such architectural gaps hinder effective remediation, as ad hoc mitigations may inadvertently cause further system degradation ( 3 ) . Recent advancements in Large Language Models (LLMs) ( 55 ) suggest new possibilities for reasoning over and reconciling architectural gaps. By leveraging the ability to process multimodal inputs, such as architecture and network topology diagrams ( 51 ) , LLMs exhibit the potential to reconstruct architectural contexts from fragmented data. A coherent architectural representation, if reliably established, could in turn provide the structural foundation for downstream security workflows, including continuous threat modelling (e.g., STRIDE-LM ( 33 ) ), dynamic attack tree synthesis ( 40 ) , and adaptive risk assessment across the CPS life-cycle. However, the systematic integration of these capabilities into a unified, architecture-centric CPS security assessment workflow remains underexplored. Figure 1 . Overview of ASTRAL. The security assessment workflow for CPSs with incomplete architectural knowledge, enabled by multimodal LLMs, progresses through three phases: architectural reconstruction (Phase 1); security modelling (Phase 2); and probabilistic analysis (Phase 3). Workflow diagram of ASTRAL framework CPS security assessment workflow using LLMss. ASTRAL ( A rchitecture - Centric S ecurity T hreat R isk A ssessment using L LMs), a multimodal LLM - driven technique for architecture - centric security assessment of CPSs, operationalises these capabilities via the multi-phase workflow illustrated in Figure 1 . ASTRAL supports practitioners in reconstructing and subsequently analysing these architectures when documentation is incomplete or absent. In this context, reconstruction refers to the systematic derivation of a coherent architectural representation from fragmented and heterogeneous artefacts. The methodology integrates prompt chaining, few - shot learning, and structured LLM reasoning ( 37 ) to systematically extract and organise architectural knowledge for security assessment. By combining LLM reasoning with architecture-centric modelling, our research facilitates the dynamic identification of threats alongside quantitative risk assessment within a unified, life-cycle-integrated framework. 1.1. Research Questions To evaluate the effectiveness and practical applicability of our proposed approach, we investigate the following research questions: • RQ1: To what extent do ASTRAL’s key design elements contribute to architectural reconstruction quality and structural validity? This research question evaluates how ASTRAL’s key design elements (multimodality, guardrails, sample configuration settings) influence the quality and structural validity of reconstructed architectures. We conduct ablation experiments to quantify the relative degradation in trust boundary and attack path validity, graph connectivity, and semantic coherence when each element is selectively disabled or modified. • RQ2: What architectural information gaps can ASTRAL fill? This research question examines the extent to which ASTRAL can address incomplete architectural information using both textual data and images. The findings will demonstrate how LLM - assisted reconstruction of architectural knowledge can help bridge the information gaps between threat modelling, attack modelling, and risk analysis stages within the security assessment workflow. • RQ3: How do cybersecurity practitioners perceive the trustworthiness and reliability of the security assessment results generated by ASTRAL? This research question explores the degree to which experienced cybersecurity practitioners perceive LLM - generated results as trustworthy and reliable. The findings will provide insight into the credibility, consistency, and correctness of LLM - assisted security assessments in CPS contexts. • RQ4: How useful is our proposed approach for CPS practitioners? This research question assesses the extent to which practitioners view LLM - assisted security assessment approaches as useful to their work. The results will help determine the practical applicability, perceived benefits, and adoption potential of ASTRAL for real - world CPS security assessments. Figure 2 . Prototype implementation of ASTRAL (left) with architectural information reconstructed from incomplete artefacts, e.g., data flow diagrams (DFDs) (right). Source code and demo available at ( 42 ) . 1.2. Contributions The main contributions of our work are: • Addressing architectural incompleteness as a core CPS challenge : We advance the understanding of how incomplete architectural knowledge impacts CPS security assessment, and propose applied methods that integrate information processing, system architecture, and risk analysis to address security challenges common across CPSs. We integrate threat and attack models directly into a Bayesian Network (BN) framework to evaluate critical CPS assets, vulnerabilities, and hazards within specific

Read the complete text