4 min readNew DelhiJan 27, 2026 10:46 AM IST
First published on: Jan 25, 2026 at 04:54 PM IST
Amid growing fears of AI slop in scientific research, over 51 papers accepted at the prestigious Conference on Neural Information Processing Systems (NeurIPS) were found to contain fake, AI-generated citations.
More than 100 hallucinated citations were found across 51 papers accepted at the annual conference, which has emerged as one of the biggest research events in the field of artificial intelligence and machine learning (AI/ML), according to a report published by AI detection startup GPTZero.
The company said it scanned a total of 4,841 research papers accepted at NeurIPS 2025 held in December last year in San Diego California, United States, for both hallucinated citations as well as AI-generated text. While 51 out of 4,841 research papers with fake, AI-generated sources is not statistically significant, NeurIPS LLM policy considers any hallucinated citations to be grounds for a paper’s rejection or revocation.
“These NeurIPS papers have already been accepted, presented live, and effectively published. Since NeurIPS 2025 had an acceptance rate for main track papers of 24.52%, each of these papers beat out 15,000 other papers despite containing one or more hallucinations,” GPTZero said in a blog post published on Wednesday, January 21.
Since NeurIPS is a gathering of some of the leading minds of AI research, having a research paper accepted by the conference is significant. However, GPTZero’s findings show that even some of the world’s leading AI experts struggle to ensure that the AI tools they use provide accurate responses.
NeurIPS is not the only research conference grappling with the issue of AI-generated writing and hallucinations.
Over 50 hallucinated citations in papers under review for ICLR 2026 were detected by GPTZero in December last year. Online pre-print repositories have been flooded with low-quality, AI-generated research papers. A recent analysis of arXiv submissions found that scientists who appeared to be using LLM-powered tools posted about 33 per cent more papers than researchers who did not use such tools, according to a report by The Atlantic.
How did GPTZero detect fake citations?
GPTZero said it used its in-house developed agentic AI tool called ‘Hallucination Check’ to scan sources cited in the more than 4,000 NeurIPS research papers and flag any citations in a document that could not be found online.
The company also added that the citations flagged by its tool as being an AI-generated fake were confirmed manually by a human. It has referred to such AI-hallucinated citations as ‘vibe citations’.
“We define a vibe citation as a citation that likely resulted from the use of generative AI. Our definition excludes obvious spelling mistakes, dead URLs, missing locators, and other errors that are plausibly human,” the company said.
GPTZero’s Hallucination Check tool has been made available for authors to check their manuscripts for citation errors — including common issues that can occur without LLM involvement like dead links or partial titles. It further said that its AI Detector tool “allows editors and conference chairs to check for AI-generated text and suspicious citations at the same time, leading to faster and more accurate editorial decisions.”
What is NeurIPS?
Founded in 1987, NeurIPS is a research conference devoted to studying neural networks and the interplay among computation, neurobiology and physics. Since neural networks underpin most of the AI systems today, the conference has evolved into a major AI event.
The 39th edition of NeurIPS held in San Diego, California, last year saw a record-breaking 26,000 attendees, twice as many as just six years ago, as per a report by CNBC.
Between 2020 and 2025, submissions to NeurIPS increased more than 220 per cent from 9,467 to 21,575. Each paper is reportedly peer-reviewed by multiple people who are instructed to flag hallucinations. Over the years, the organisers of the conference have had to recruit greater numbers of reviewers in order to continue its stated mission of “rigorous scholarly publishing in machine learning and artificial intelligence.”
Ironically, a research paper published months ahead of NeurIPS 2025, titled ‘The AI Conference Peer Review Crisis’, had warned of AI-generated, fake citations as a potential problem that could arise at the conference.
The fear of being labelled is real. Editors of literary journals and prize panels are also feeling it. (File photo) 
Share
