Beyond Hallucination: Ethical Risks and Narrative Dislocation in Generative AI

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This paper examines the ethical implications of hallucination in generative AI, particularly in narrative production contexts such as film, media, and automated content generation. While hallucination is often treated as a technical limitation, we argue that it introduces a deeper ethical risk: the production of coherent yet misleading content that disrupts the relationship between meaning, interpretation, and responsibility. We develop the concept of “narrative dislocation” to explain how hallucinated outputs generate structural inconsistencies at the production level and induce interpretive shifts at the audience level. Drawing on computational evidence from DramaBench and a synthesis of experimental studies on AI disclosure, we show that such dislocation is not incidental but structurally persistent across advanced models. The paper identifies three key ethical challenges: the erosion of epistemic trust, the ambiguity of authorship and responsibility, and the unintended harms produced by transparency-based governance mechanisms. In response, we propose a framework of “narrative ethics alignment,” which integrates evaluation standards, production processes, and institutional governance. This study contributes to AI ethics by bridging technical phenomena and normative concerns, offering a new perspective on responsibility and trust in AI-mediated communication systems.