Artificial Intelligence: Trust in Artificial Intelligence (AI) is increasing rapidly, but now this technology is raising questions on the credibility of scientific research. A recent report has revealed that in many research papers included in the world’s prestigious AI conference, such references were found which did not actually exist. These references were fabricated by AI tools which in expert parlance are called hallucinations.
A shocking case came to light in NeurIPS
Fake, AI-generated citations were found in 51 research papers accepted at NeurIPS 2025, considered the largest conference on AI and machine learning. This information came to light in the report of AI detection startup GPTZero. During the investigation of a total of 4,841 research papers, more than 100 such references were found in these 51 papers which had no real existence.
Small numbers, but big impact
Even though the figure of 51 out of more than 4,800 papers may seem statistically small, the policy of NeurIPS is very strict. Here, if even a single false citation is found in any research paper, then that paper can be rejected or cancelled. The special thing is that these papers had already been approved, they were also presented on the stage and were considered formally published.
Why are even top AI experts missing out?
Getting the paper accepted in NeurIPS is considered a big achievement in itself. Here the selection rate in the main track is around 25 percent, that is, these researches are selected after leaving behind thousands of papers. GPTZero’s report shows that even the world’s top AI researchers are not able to fully verify the accuracy of the AI ​​tools they are using.
The problem is not limited to NeurIPS only.
This problem related to AI is not limited to just one conference. More than 50 fake citations were also detected in the papers under review for ICLR 2026. Additionally, the number of low-quality, AI-generated research papers is also increasing rapidly on pre-print platforms like arXiv. According to an analysis, researchers who are using LLM based tools are submitting about 33 percent more papers than others.
How did GPTZero catch fake citations?
GPTZero used its special AI tool ‘Hallucination Check’. This tool scanned the references given in thousands of research papers and identified those citations for which no online record was found. After this, every citation identified as suspicious by AI was manually checked by humans to ensure that there was no room for error.
Also read:
WhatsApp can be hacked even without OTP! To avoid this fraud, keep these 5 important steps in mind
