VerusCite Says Springer’s AI Reference Checks Missed Hallucinated Citations

Summary

VerusCite argues that Springer Nature’s internal AI tool for checking citation relevance is not reliably detecting hallucinated references. The author distinguishes relevance checking from verifying whether a cited source actually exists, while questioning how a hallucinated reference can be judged relevant. VerusCite identified suspected hallucinated citations in ten articles published by Springer Nature and other major publishers in August and September 2026, and says each finding was confirmed through a second human check. The examples include papers on large language models, generative-AI bias, LLM evaluation, education, crime, violence, and machine learning. The article says similar problems appear across Taylor & Francis, Sage, Oxford, Elsevier, MDPI, PLOS One, and several preprint platforms, although Wiley is not yet represented in the author’s collection. The author contends that publishers perform only limited integrity verification and leave much of the responsibility to reviewers and editors. VerusCite is presented as costing $2 per paper review, which the author uses to argue that reference-accuracy checks should be a minimal expectation for academic journals.