AI’s Unintended Consequence: Fabricated Citations at NeurIPS Conference
In a recent analysis, AI detection startup GPTZero examined 4,841 papers accepted at the prestigious Conference on Neural Information Processing Systems (NeurIPS) held in San Diego. The company identified 100 fabricated citations across 51 papers, highlighting an ironic twist in the AI research community.
Acceptance into NeurIPS is a significant achievement in the AI field. Given the prominence of the conference, it’s reasonable to assume that authors might utilize Large Language Models (LLMs) to streamline the citation process. However, the discovery of these hallucinated citations—references to non-existent or incorrect sources—raises concerns about the reliability of AI-generated content.
While 100 fabricated citations among tens of thousands may seem statistically insignificant, the issue underscores a broader challenge. NeurIPS emphasizes rigorous scholarly publishing in machine learning and artificial intelligence, with each paper undergoing multiple peer reviews designed to detect inaccuracies. Despite these measures, some AI-generated errors slipped through, suggesting that even the most meticulous processes can be vulnerable.
Citations serve as a metric of a researcher’s influence and credibility. When AI tools generate false references, it undermines the integrity of academic work and the trustworthiness of scholarly communication.
The peer review process, already burdened by a surge in submissions, faces additional strain from AI-generated content. GPTZero’s findings point to a submission tsunami that has overwhelmed review pipelines, making it challenging to catch every fabricated citation. This situation was highlighted in a May 2025 paper titled The AI Conference Peer Review Crisis, which discussed the pressures on premier conferences like NeurIPS.
The responsibility also lies with researchers to verify the accuracy of their citations. Relying solely on AI tools without thorough fact-checking can lead to the dissemination of false information, compromising the quality of research.
This incident serves as a cautionary tale about the limitations of AI in academic writing. While AI can assist in drafting and organizing content, human oversight remains crucial to ensure accuracy and reliability.
In conclusion, the discovery of fabricated citations at NeurIPS highlights the need for a balanced approach to integrating AI into academic research. It emphasizes the importance of human diligence in maintaining the integrity of scholarly work in the age of artificial intelligence.
