Epistemic Infrastructure, Generative AI, and the Governance of Digital Memory: Indexing, Power, and the Political Economy of Knowledge

Generative artificial intelligence is commonly described as a novel engine of knowledge production, yet its operation depends on pre-existing infrastructures through which texts and records become discoverable, contextualised, and machine-readable. This article asks: how should archival and metadata infrastructures be conceptualised and governed when they contribute to AI-mediated knowledge production but remain institutionally fragile and only partly visible in copyright and AI policy? Using a conceptual, interdisciplinary analysis of scholarship in science and technology studies, archival and memory studies, political economy, and law, it compares the United States, United Kingdom, and European Union and uses the 2023 British Library cyberattack as an illustrative case of infrastructural vulnerability. The article makes three contributions. It conceptualises archival description and metadata as epistemic infrastructure; identifies a governance gap between copyright, database rights, and AI transparency rules; and develops a Digital Memory Charter with five implementable commitments: infrastructural transparency, reciprocal benefit, metadata stewardship, democratic oversight, and pedagogical responsibility. The analysis does not claim that archival metadata has a uniform causal effect on model training or that particular commercial models used British Library holdings where public evidence is unavailable. It instead establishes a defensible basis for governing documented uses of cultural-heritage corpora and for improving transparency where those uses cannot presently be verified.