Hermeneutic Sovereignty under Generative AI: A Humanities Framework for Protecting Human Meaning-Making from Interpretive Foreclosure

Abstract

Generative artificial intelligence increasingly mediates not only what institutions decide, but how institutions interpret people. Large language models draft case summaries, student feedback, performance reviews, clinical notes, policy briefs, creative text and administrative explanations. Existing AI governance frameworks appropriately emphasise accuracy, fairness, transparency, accountability and human oversight, yet these categories do not fully capture a distinct humanities problem: a fluent machine-generated interpretation can become institutionally authoritative before the person concerned has meaningfully articulated, contextualised or contested the account. This paper develops a person-centred framework of hermeneutic sovereignty for generative AI. Using an interdisciplinary conceptual methodology, it synthesises philosophical hermeneutics, epistemic injustice, narrative identity, human factors research, empirical studies of generative AI and contemporary governance frameworks. Hermeneutic sovereignty is defined as the situated and relational standing and capacity of persons and communities to participate meaningfully in constructing, contextualising, contesting, pluralising, revising and, where appropriate, refusing interpretations of their own experiences, identities, intentions, reasons and circumstances when those interpretations are mediated by AI. The paper identifies four mechanisms of AI-mediated interpretive foreclosure: interpretive pre-emption, hermeneutic compression, authority laundering and recursive fixation. It then proposes six dimensions for evaluating hermeneutic sovereignty and an Interpretive Foreclosure Test for institutional workflows. The analysis shows that human oversight is insufficient when the human merely approves a machine-framed account. Responsible adoption requires governance of the meaning-making process itself, including human-first elicitation, source-to-summary traceability, visible uncertainty, plural framing, contestability, correction propagation and time-bounded interpretive records. The paper concludes that trustworthy AI must preserve not only a human role in decisions, but a meaningful human standing in the production of the interpretations on which decisions depend.