Treat AI-Hallucinated Citations as Research Misconduct

A new correspondence published in Nature on 27 October 2026 argues that citations fabricated by artificial-intelligence tools should be treated as research misconduct. The letter, written by Jan Pisek of the University of Tartu in Estonia, responds to recent reports that the preprint server arXiv has begun rejecting papers containing references hallucinated by AI systems. The piece enters an ongoing scientific debate over whether AI-generated false citations are merely editorial failures or constitute violations of research integrity norms. The correspondence comes as research institutions and publishing platforms consider how to respond to the growing use of generative AI in scientific writing. arXiv’s moderation move reflects broader concern across preprint platforms about so-called AI slop, where low-quality or fabricated submissions can distort scientific literature, waste reviewer time, and undermine public trust. Researchers remain divided on responsibility, particularly in cases where AI tools assist drafting or literature organization without explicit human verification of every citation. By framing hallucinated references as misconduct, Pisek’s letter supports stronger enforcement mechanisms, including clearer author obligations to validate AI-assisted outputs before submission. The argument also aligns with wider calls for stricter editorial controls, independent scrutiny of AI-assisted scholarship, and better disclosure practices as large language models become more embedded in academic workflows.