Artificial intelligence in archives: six modes of involvement and an ecosystem of trust — a reading in light of the theoretical-methodological framework of InterPARES

Abstract Artificial Intelligence (AI) does not constitute a single phenomenon in archival practice. It may generate potential records, support archival functions, assist tool development, analyze holdings, become an object of preservation, or mediate access. Treating these different modes of involvement as homogeneous produces imprecise diagnoses and inadequate safeguards. This theoretical-conceptual article proposes a six-mode model: AI-assisted record production, AI tools for archival functions, AI-assisted tool development, AI analysis of archival holdings, AI systems as preservation objects, and AI-mediated access. Each mode is evaluated against common requirements of archival trust — authenticity, reliability, accuracy, and the chain of preservation — and is subject to a cross-cutting governance framework based on paradata, explainable AI, bias mitigation, ethical use, and accountable human oversight. The analysis is guided by the theoretical-methodological framework of InterPARES, particularly InterPARES Trust AI (2021–2026). The article concludes that AI can support archival work only when its use is governed by archival principles, documented procedures, and clearly assigned human and institutional responsibility. The model provides an instrument for analyzing AI initiatives, designing safeguards, and guiding future empirical validation through case studies.