Trustworthy AI for high-stakes decision support across critical sectors

Artificial intelligence (AI) is increasingly embedded in high‑stakes decision support across domains ranging from cybersecurity and energy management to healthcare, finance and public governance. Ensuring trustworthiness is essential, yet evidence remains fragmented across sectors. This title‑driven scoping review synthesises recent publications that investigate AI‑enabled decision support under conditions of criticality and risk. Following a portfolio‑bounded methodology, the review infers from titles a cross‑sector taxonomy of trust dimensions—including interpretability, robustness, privacy, fairness and governance—and maps them onto sector‑specific tasks such as threat detection in critical infrastructure, optimisation of renewable energy systems, diagnosis and prognosis in healthcare, fraud detection in digital finance, and policy compliance in welfare management. The extraction schema codes sectors, tasks, methodological families and deployment settings based solely on title information, and a conceptual rigor rubric is proposed to appraise future full‑text evidence. Narrative synthesis highlights common design principles such as interpretable‑by‑design models and privacy‑preserving federated learning, recurring failure modes such as dataset bias and adversarial vulnerabilities, and deployment considerations spanning edge, cloud, 6G and management information systems. A research agenda is outlined to guide systematic evidence gathering, with priorities for unified trust metrics, cross‑sector federated collaboration, human‑centric explainability, adversarial resilience and governance integration. Limitations due to title‑only inference are acknowledged, and the review serves as a structured foundation for future evidence‑based synthesis.