The escalating complexity and sophistication of cyber threats necessitate advanced solutions that not only counteract these threats but also proactively adapt to the evolving needs of diverse stakeholders involved in digital infrastructures, such as telecom operators, cloud service providers, and end-users in sectors like healthcare and finance. This research addresses a crucial gap by focusing on a systemic, AI-powered approach to stakeholder needs analysis in cybersecurity. By aligning closely with stakeholder requirements, the proposed framework aims to offer dynamic, responsive cybersecurity solutions that enhance the resilience of digital infrastructures against evolving cyber threats. This research systematically maps the landscape of stakeholder needs in cybersecurity across different sectors through qualitative methods like interviews and focus groups, supplemented by data from the CyberSecDome project’s pilot cases and open calls. Requirements for an AI-driven framework are then formulated based on these data to identify patterns and predict stakeholder needs. The analysis reveals critical challenges faced by stakeholders, including limited threat intelligence sharing, insufficient automation in incident response, and regulatory hurdles related to data protection laws and evolving cybersecurity legislation. There is a strong interest in leveraging AI for enhanced intrusion detection, real-time threat intelligence sharing, and privacy-preserving information exchange.

Cybersecurity risk is one of the primary and growing concerns for ensuing security and resilience of organizations, regardless of their size and type. While proactive risk management is effective, it is challenging due to the evolving and sophisticated threat landscape, exploitation of known and unknown vulnerabilities, and a dynamic security context. The dynamic security context further complicates to calculate the accurate risk level, leading to risk perception that can vary between different stakeholders. However, the demand for adopting cyber insurance is increasing as an effective risk mitigation strategy to avoid any potential loss. In this context, this paper proposes an Explainable AI (XAI) based dynamic cybersecurity risk management approach for informed cyber insurability decision making. The approach utilizes an Large Language Model (LLM) based framework for real-time, contextualized risk level assessment and adopts XAI techniques such as feature contribution and correlation, to justify the decision making. A comprehensive evaluation using an industrial use case and experiment demonstrates the applicability of the proposed approach. The experiment part uses a widely used vulnerability dataset to predicate high exploitable vulnerabilities and links them with the identified assets of the use case scenario. The result shows 96.9% accuracy for the exploitable vulnerability identification and XAI operationalisation justifies the selection of right security control and the cyber insurability decision based on the residual risk.

System-to-system communication via Application Programming Interfaces (APIs) plays a pivotal role in the seamless interaction among software applications and systems for efficient and automated service delivery. APIs facilitate the exchange of data and functionalities across diverse platforms, enhancing operational efficiency and user experience. However, this also introduces potential vulnerabilities that attackers can exploit to compromise system security, highlighting the importance of identifying and mitigating associated security risks. By examining the weaknesses inherent in these APIs using security open-intelligence catalogues like CWE and CAPEC and implementing controls from NIST SP 800-53, organizations can significantly enhance their security posture, safeguarding their data and systems against potential threats. However, this task is challenging due to evolving threats and vulnerabilities. Additionally, it is challenging to analyse threats given the large volume of traffic generated from API calls. This work contributes to tackling this challenge and makes a novel contribution to managing threats within system-to-system communication through API calls. It introduces an integrated architecture that combines deep-learning models, i.e., ANN and MLP, for effective threat detection from large API call datasets. The identified threats are analysed to determine suitable mitigations for improving overall resilience. Furthermore, this work introduces transparency obligation practices for the entire AI life cycle, from dataset preprocessing to model performance evaluation, including data and methodological transparency and SHapley Additive exPlanations (SHAP) analysis, so that AI models are understandable by all user groups. The proposed methodology was validated through an experiment using the Windows PE Malware API dataset, achieving an average detection accuracy of 88%. The outcomes from the experiments are summarized to provide a list of key features, such as FindResourceExA and NtClose, which are linked with potential weaknesses and related threats, in order to identify accurate control actions to manage the threats.

AI-enabled applications have achieved widespread adoption for complex problem-solving and informed decision-making. However, growing concerns regarding AI system failures that lead to bias, inequalities, and untrustworthy outcomes necessitate a move beyond performance evaluation to ensure trustworthy, equitable decisions. Although existing practices identify key trustworthiness characteristics, they lack a unified pipeline for systematically aligning these traits across the development lifecycle. To address this gap, this work presents a novel holistic Trustworthy AI (T-AI) pipeline that integrates six key characteristics—privacy, fairness, security, robustness, safety, and explainability—into specific development phases. To formally govern this integration, we introduce a novel Adaptive Trustworthiness Integration (ATI) algorithm, which enforces an irreversibility-based ordering of characteristics and applies conditional gating at each phase boundary to prevent propagation of unresolved violations, producing an empirically measured inter-characteristic trade-off matrix that makes intervention costs and benefits transparent and reproducible. The pipeline is validated through two experiments: an industrial pilot at Fraunhofer IBMT, extended with multi-seed validation across three seeds and a large-scale public microscopy benchmark (PanNuke; 7,904 images), achieving 99.5% mAP@50; and a healthcare experiment using the Synthea dataset, maintaining 89.5% predictive accuracy. These results demonstrate that the pipeline reduces demographic fairness gaps from 46% to under 5%, achieves a 0% rate of clinically dangerous hallucinations via a four-layer safety architecture, and confirms consistent trustworthiness guarantees across datasets and random initialisations. Ultimately, this work advances United Nations Sustainable Development Goal 9 by ensuring fair and privacy-preserving data collection, preventing system failures through adversarial hardening, and enabling transparent decision-making.