Audit Finds Suspected Fabricated References in Science, Engineering, Computing

A Columbia University-led research team has found suspected fabricated references across physics, chemistry, materials science, engineering and computing, extending earlier medical literature audits that identified similar integrity problems. The new study, published in the Proceedings of the National Academy of Sciences, analyzed 500,000 science, engineering and computing papers indexed in OpenAlex and reviewed 21.6 million references carrying digital object identifiers. The automated verification tool flagged 1,135 references in 619 papers as fabricated because the cited titles could not be found online. In a random sample of 400 flagged references reviewed by two independent experts, 85.8 percent were confirmed as fabricated. The study found the problem in every discipline examined, with rates of 25.6 suspected fabricated references per 10,000 papers in physics, 18.4 in materials science, 11.1 in computer science and artificial intelligence, 10.3 in engineering and 4.3 in chemistry. The audit also detected an accelerating trend. The share of affected papers rose from about 1 in 1,600 in 2023 to 1 in 650 in 2025 and 1 in 435 in the first quarter of 2026. Most affected papers contained one or two fabricated references, but 27 papers contained five or more, accounting for more than a third of all suspected fabricated references identified. Two of the 619 affected papers had been retracted at the time of the audit. Applying the observed rates to roughly 7 million papers in the examined fields, the authors estimate that about 8,500 published papers may contain at least one fabricated reference. Maxim Topaz, an associate professor at Columbia Nursing and Columbia University Data Science Institute who led the study, said the problem is not specific to medicine. He emphasized that references create an expectation that cited sources exist and said automated screening before publication could help editors detect the issue. The researchers recommend routine automated reference checks at submission, retroactive screening of published literature, and separate tracking of fabricated references in indexing services and research-integrity databases. The findings build on the team’s earlier work. A May research letter in The Lancet identified 4,046 fabricated citations in 2,810 of 2.5 million medical papers. A July paper in the Journal of Internal Medicine found fabricated references were more common in medRxiv preprints than in peer-reviewed articles. A September arXiv preprint reported that 26 large language models fabricated 55 percent of medical references they were asked to provide. The researchers are next screening up to 12 million medical papers published since 2020 and are working with journal editors and research integrity officers to develop submission screening protocols, policies and training.