Large language models in cyber humanities: companionship, stewardship and repeatability

Purpose Drawing from recent advancements, this study aims to highlight how the increasing integration of large language models (LLMs) necessitates a novel governance scheme regarding data integration, ethical oversight and bias mitigation. Design/methodology/approach Central to the scheme are the concepts of LLMs companionship and stewardship, which ensure responsible governance to promote repeatability and robust assessment. Findings This study proposes an operational scheme for librarian-led stewardship of LLM-supported research, detailing governance roles, data curation workflows, and accountability mechanisms for attribution and dating tasks. Research limitations/implications This paper offers a conceptual viewpoint rather than large-scale empirical validation. Its framework relies on an illustrative case study and conceptual reasoning rather than extensive implementation data. The author invites the digital library and cyber humanities (CH) communities to test, adapt and empirically evaluate the proposed governance practices. Practical implications Library-based AI governance has practical implications in managing the relationship between human CH researchers and LLMs. Practical policy adjustments are required to position librarians as active stewards of ethical, repeatable and provenance-rich LLM-assisted research. Social implications By positioning libraries as ethical stewards of LLM-supported CH research, this framework advances epistemic justice, mitigates cultural and algorithmic biases against marginalized voices, promotes inclusive representations of global heritage and strengthens public trust in transparent, equitable and democratized knowledge production in the AI era. Originality/value The scheme advocates library-based AI governance to address challenges in the relationship between human researchers and LLMs in humanities scholarship, positioning librarians as active stewards of ethical, repeatable and provenance-rich computational research.