From instrument to institution: generative AI and the epistemic dispossession of the academic library

Purpose This paper analyses the emergence of generative artificial intelligence (AI) not as a further tool in the automation of search but as an epistemic institution that produces, validates and distributes knowledge. It examines the structural mechanisms through which this transition marginalises academic libraries, displacing them from the epistemic functions of synthesis, evaluation and curation they have historically performed. Against the assumption that marginalisation is a technological inevitability, the paper argues that it is conditional and articulates the institutional conditions under which academic libraries can recover epistemic agency and reposition themselves as active participants in the knowledge infrastructure of the AI era. Design/methodology/approach The paper adopts a conceptual, abductive approach, employing the Tron film trilogy as an analytical model of epistemic infrastructure transition rather than as a cultural metaphor. It integrates Science and Technology Studies, the philosophy of technology and library and information science scholarship. The historical reconstruction proceeds inductively, deriving three phases of library infrastructure development from particular technologies, from card catalogues to commercial scholarly platforms. The diagnosis proceeds deductively, deriving specific mechanisms of marginalisation from the premise that generative AI internalises epistemic functions. The recovery programme then translates that diagnosis into the institutional conditions under which academic libraries can respond. Findings Generative AI constitutes a qualitative break in the history of knowledge infrastructure, crossing from the category of retrieval tool to that of epistemic institution that organises the conditions under which knowledge is produced and validated. Three mechanisms drive library marginalisation: disintermediation of the institutional mediator, dependency on training infrastructures beyond library governance and the transfer of epistemic authority from accountable institutions to opaque algorithmic systems. Scenario analysis yields three possible futures, full marginalisation, coexistence and co-authorship of epistemic infrastructure. The third is not inevitable but contingent on libraries claiming roles in data stewardship, AI governance and critical AI literacy. Research limitations/implications As a conceptual contribution, the paper develops an analytical framework rather than empirical findings; its three mechanisms and three scenarios are theoretically derived and await empirical testing. The evidence base draws substantially on Anglophone, predominantly US LIS scholarship, which may limit transferability to other national and institutional contexts. Future research should operationalise and empirically validate the proposed mechanisms of marginalisation, assess the conditions for epistemic recovery across diverse library settings and examine longitudinally whether libraries that invest in data stewardship, AI governance and critical AI literacy realise the co-authorship scenario the framework identifies. Practical implications Academic libraries should pursue three mutually reinforcing strategies. First, they should reposition themselves as active participants in the governance of the data ecosystems on which AI systems are trained, extending research data management towards stewardship of training corpora. Second, they should secure inclusion in university AI governance structures, from which information professionals are frequently absent despite directly relevant competencies in information ethics, source evaluation and quality assurance. Third, they should reconceptualise critical AI literacy, the capacity to evaluate not only what AI produces but how, as a defining institutional mission rather than a supplementary digital skill. Social implications If generative AI becomes the default mediator of scholarly knowledge, the consequences extend beyond libraries to the epistemic commons on which science, scholarship and democratic deliberation depend. Unaccountable, non-transparent knowledge production risks normalising narrow epistemic positions, eroding epistemic pluralism and widening inequalities in knowledge access already stratified by institutional wealth. Positioning academic libraries as public-interest guarantors of accountable curation, epistemic equity and critical AI literacy helps safeguard citizens' capacity to evaluate AI-generated knowledge, sustains diverse knowledge traditions and protects the reliability of the shared information environment on which informed public life increasingly rests. Originality/value The paper offers a novel integrative framework that combines Science and Technology Studies, the epistemology of AI and library and information science scholarship. Its central contribution is to reframe library marginalisation not as a contingent service challenge to be met by incremental adaptation, but as a structural consequence of AI's emergence as an epistemic institution. Methodologically, it advances the use of the Tron trilogy as an analytical model rather than an illustrative metaphor. It provides one of the first systematic articulations of academic libraries as potential co-architects, rather than residual beneficiaries, of the knowledge infrastructure of the AI era.