Borrowed Voices, Shared Debt: Plagiarism, Idea Recombination, and the Knowledge Commons in Large Language Models

Large language models generate fluent text by recombining the language and ideas of prior authors at scale. This process produces plagiarism-like harms in three dimensions: direct wording leakage, imitation of distinctive styles, and appropriation of argument structures or conceptual syntheses without provenance. At the same time, their capacity to provide insight or novel-seeming combinations depends entirely on the accumulated labor of millions of human writers, editors, teachers, and curators who built the knowledge commons. This paper argues that denunciation and recognition must proceed together: the harms of extraction must be exposed, yet the debt to the commons must also be acknowledged. The article proposes a framework that defines the scope of plagiarism in this context, diagnoses the mechanisms of recombination, and sets out operational remedies, including dataset governance, attribution layers, compensation pools, and measurable audit thresholds. The goal is to establish a system that restricts illegitimate appropriation while reinvesting in the infrastructures of shared knowledge that make such synthesis possible.DOIPrimary archive: https://doi.org/10.5281/zenodo.17132004Secondary archive: https://doi.org/10.6084/m9.figshare.30137422SSRN: Pending assignment (ETA: Q3 2025)