## AI Hallucinations Contaminate Scientific Studies GPTZero, a system for detecting AI-generated content, has discovered that several studies presented at the NeurIPS conference include citations produced by language models. Specifically, approximately 100 suspicious citations were identified in 51 publications. This phenomenon highlights a growing problem: researchers are increasingly relying on AI assistance, but without adequately verifying the accuracy of the information provided. The integrity of scientific research could be compromised if more rigorous measures are not taken to validate the results obtained with the aid of artificial intelligence tools. ## Implications and Future Perspectives The use of AI in scientific research offers undoubted advantages, such as the automation of repetitive tasks and the analysis of large amounts of data. However, it is essential that researchers maintain a critical approach and carefully verify the results obtained. The adoption of clear guidelines and effective control tools can help mitigate the risks associated with the use of AI and preserve the reliability of science.