The Intelligent Archival Ecosystem: Designing the Future Architecture of AI-Powered Archives

Purpose: This article reviews how artificial intelligence (AI) technologies are reshaping the theory and practice of archival science and, on the basis of that review, proposes a conceptual reference architecture for what is termed an “intelligent archival ecosystem” – an integrated model in which AI-based tools support, rather than replace, the professional judgement of archivists across the record lifecycle.Method/approach: The study combines three components: (1) a structured narrative review of scholarly, professional and policy literature on AI in archives, records management and digital preservation, drawing mainly on sources published between 2019 and 2026, including prior systematic reviews, practitioner studies, standards and official guidance from national archives, together with a cross-study analytical synthesis of eleven empirical evaluations; (2) a documentary analysis of six publicly reported implementations and policy responses, used to test and refine the architecture; and (3) design reasoning grounded in archival theory, including the records continuum model, the OAIS reference model (ISO 14721:2025), Records in Contexts and ISO 15489.Results: The review shows that natural language processing and computer vision are the most mature techniques; that accuracy depends strongly on the interpretive demands of the task (in one national-library experiment, up to 99% for title and author fields but below 50% for subject and genre); and that the provenance of AI-generated metadata and governance mechanisms are rarely evaluated in published implementations. The article proposes a five-layer architecture – ingestion and capture, AI-augmented description and appraisal, trust and provenance, discovery and access, and governance – that situates natural language processing, computer vision, machine-learning classification and large language models within a human-in-the-loop framework accountable to the archival principles of authenticity, reliability, integrity and usability, and it introduces field-level automation tiers as a design principle. Experimental and interview evidence indicates that recognition and description quality depend on adaptation to the script and the collection, and that hands-on exposure to imperfect AI output lowers participants’ trust and willingness to use AI. A protocol for empirical validation through multiple case studies is specified.Conclusions/findings: AI can substantially increase the efficiency and accessibility of archival description, appraisal and retrieval, but its responsible integration requires explicit governance structures, transparency mechanisms, the treatment of AI-related materials as records in their own right, and a redefined professional role for the archivist as orchestrator and auditor of automated processes rather than as sole executor of archival functions. Because the model is conceptual and evidence-informed rather than empirically validated, the article identifies multi-institutional case-study research as the priority for future work.