Towards Trustworthy Multi-Agent Cybersecurity Architectures for Financial Critical Infrastructures
Year: 2026
Type: preprint
Abstract: Financial critical infrastructures face mounting cyber threats across heterogeneous, fragmented digital ecosystems. Existing Security Operations Centre architectures, spanning SIEM platforms, XDR systems, and threat intelligence feeds, lack the contextual reasoning, governance alignment, and explainability that highly regulated environments require. We propose a trustworthy multi-agent cybersecurity architecture that integrates LLM-enhanced contextual orchestration grounded in shared semantic security knowledge, human-in-the-loop governance, compliance-aware runtime control, and collaborative threat intelligence. Rather than treating cyber resilience as a purely technical detection problem, our framework conceptualises it as a socio-technical orchestration challenge explicitly designed to support compliance with DORA, NIS2, GDPR, PSD2, and the EU AI Act. We describe a six-layer architecture, analyse its governance and resilience implications, and identify open research challenges. The central argument is that AI agents should augment cybersecurity professionals, not replace them, preserving human authority over high-stakes decisions while reducing alert fatigue and enabling privacy-preserving collaborative defence across institutions.
Source: Preprints.org
Authors: Fernando Luis-Ferreira, Carla Loureiro-Rodrigues, John Soldatos, Elina Vasiliki Papadopoulou, Stamatia Moniou, Pedro Maló, George Kousiouris, Panagiotis Rizomiliotis, Louiza Kachrimani, Thanassis Giannetsos, Nikos Papageorgopoulos, Ernstjan de Gooyert, Stefan Fürnsinn, Konstantina Tripodi, Ioannis Chouchoulis
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7234801408
