The transformation of human-authored textual corpora into predictive generative capacity without transparent source attribution or recoverable provenance. The paper shifts the AI plagiarism debate from copying and memorization toward structural appropriation, recombinative authorship, and generative provenance. Large language models introduce a form of plagiarism that cannot be reduced to verbatim copying or copyright infringement. Their central operation is structural appropriation: the absorption, recombination, and redeployment of human intellectual labor under conditions of referential opacity and attribution collapse. Structural appropriation; Recombinative plagiarism; Referential opacity; Attribution collapse; Synthetic originality; Predictive authorship; Latent intellectual debt; Corpus parasitism; Invisible intellectual labor; Generative provenance.