For most of literary history, the question of who wrote a text carried an assumption so obvious it rarely needed stating: a person did. The fluency of contemporary large language models has unsettled that assumption, producing sonnets, short stories, and novel chapters that readers in controlled studies frequently cannot distinguish from human-authored work. This paper argues that AI-generated literature does not so much destroy literary authenticity as force a reconfiguration of what the term has always meant. Synthesizing literary theory, empirical reception research, copyright law, and consumer psychology, the paper traces four converging lines of evidence: readers judged blind often cannot detect machine authorship and sometimes rate it more favorably, yet the same readers penalize a text emotionally once its AI origin is disclosed; structural analyses show AI-generated fiction and poetry remain measurably less inventive than the best human work even when formally polished; and copyright authorities have sidestepped the detection problem entirely by anchoring protection to traceable human decision-making rather than textual quality. Revisiting Roland Barthes's "death of the author" alongside a 2016 case in which an AI-assisted novella nearly won a Japanese literary prize, the paper argues that authenticity was never a property readers detected inside a text but a social practice negotiated among readers, critics, publishers, and legal institutions. It concludes that a more defensible, procedural model of authenticity is emerging  one grounded in disclosed, traceable human judgment rather than the simple fact of a byline.

Keywords : AI-generated literature; authorship; authenticity; large language models; Roland Barthes; computational creativity; copyright law; literary reception studies