Researchers at GPTZero have uncovered over 100 instances of hallucinated citations within papers accepted at the NeurIPS 2025 conference. These fabricated references, missed by multiple reviewers, point to a growing problem of AI-generated content and potential academic fraud. The findings underscore the strain on peer-review processes, as conferences struggle to keep pace with the influx of submissions.

The issue stems from the increasing pressure to publish and the rise of tools that generate academic-sounding text. Conferences like NeurIPS, with its high prestige, attract thousands of submissions annually. The sheer volume makes it harder to thoroughly vet each paper. The NeurIPS LLM policy states that hallucinated citations are grounds for rejection, but the scale of the problem is now clear.

The report highlights specific examples of fabricated references, including fake authors, titles, and publication details. With an acceptance rate of just 24.52%, each of these papers beat out thousands of others. This raises serious questions about the integrity of the peer-review process and the potential for the spread of misinformation in scientific literature.

Going forward, expect more scrutiny of published papers and perhaps, changes to how conferences review submissions. Tools like GPTZero are becoming essential to detect AI-generated content and ensure the accuracy of scientific research. The focus will likely shift towards more rigorous verification methods and improved detection of fabricated sources.