Artificial Intelligence in Construction Health and Safety Management Systems: An Integrated Framework for Context, Coherence, Accountability and Preventive Performance

Abstract

Artificial Intelligence in Construction Health and Safety Management Systems examines how artificial intelligence can contribute to occupational health and safety (H&S) management in construction beyond the technical performance of individual applications. Drawing on ISO 45001, HSG65 and an evidence-informed analysis of emerging AI applications, the monograph develops an integrated conceptual framework for examining AI-supported information, decisions and actions within construction H&S management systems. The research addresses five interconnected Research Problem Areas (RPAs): RPA1 — Contextual Validity: The adequacy and currency of information and context supporting AI-assisted H&S functions. RPA2 — System Integration and Coherence: The consistency, traceability and interaction of AI-supported information across H&S processes. RPA3 — Accountable Delegation: The boundaries of AI-supported decision authority and human and organisational accountability. RPA4 — Safety Signals and Organisational Learning: The relationship between AI-supported detection, corrective action, preventive effectiveness and learning. RPA5 — Observability and Evidence Distribution: The uneven availability and visibility of evidence across AI applications and H&S management functions. The monograph integrates these research areas to distinguish technical performance, field performance, preventive effectiveness and organisational value, while examining the transversal role of professional judgement, human–AI interaction and organisational governance. The work is a conceptual and evidence-informed contribution to the independent research programme on Artificial Intelligence, Construction Health & Safety and Risk Management. It does not claim empirical validation of the proposed integrated framework. Preprint — Version 1.0 — 10 October 2026. Access: The bibliographic record and DOI are publicly available. Access to the full-text PDF is restricted and may be requested through Zenodo.