A new audit conducted by researchers at Columbia University has uncovered that nearly 3,000 peer-reviewed medical papers contain fabricated citations that do not exist in scientific databases. The study, titled Fabricated citations: an audit across 2·5 million biomedical papers, was published in The Lancet and highlights a rapidly escalating crisis in academic integrity driven by the widespread adoption of AI writing tools. Led by Dr. Maxim Topaz, an associate professor at the Columbia University School of Nursing and Data Science Institute, the team developed an automated AI verification system to analyze 2.5 million open-access papers indexed in PubMed Central between January 1, 2023, and February 18, 2026. The audit examined 97.1 million references and identified 4,046 fake citations distributed across 2,810 articles. The findings reveal a dramatic surge in citation fraud. While the rate of fabrication remained relatively stable in 2023, with approximately four fake citations per 10,000 papers, the rate jumped sharply starting in mid-2024. By early 2026, the fabrication rate had climbed to approximately 57 per 10,000 papers, representing a more than twelve-fold increase. The implications of this discovery extend far beyond academic records. Dr. Topaz emphasized that medical professionals rely on clinical guidelines and evidence-based research to make treatment decisions for patients. When these guidelines are built on non-existent references, patient care is directly compromised. The audit found instances where the majority of a paper's references were fabricated; one reviewed paper contained 18 fake references out of 30. Furthermore, many of these false citations have already been propagated, appearing in other papers and systematic reviews that inform clinical care. Despite the scale of the issue, accountability has been minimal. At the time the data was collected, 98.4% of the affected papers had received no action from their respective publishers. In response, the authors have issued several recommendations to the scientific community. They urge publishers to implement verification checks for every reference included in new submissions. Additionally, they propose that indexing services add specific metadata to records to allow users to assess the accuracy of citations. The team also calls on major research integrity databases to create a dedicated category for fake references to enable systematic tracking and enforcement. Furthermore, the researchers recommend that publishers retroactively screen existing publications and issue corrections or retractions for papers where fake references undermine the conclusions. The study also includes an accompanying commentary by Dr. Howard Bauchner and Dr. Frederick P. Rivara, who stress that public trust in science is already declining globally. They argue that this crisis underscores the urgent need to maintain and improve research integrity, stating that authors must take full responsibility for the entire content of their manuscripts, including their reference lists.