Researchers audited biomedical literature and found thousands of fabricated references embedded in scientific papers, underscoring a growing integrity risk as AI tools become routine in academic workflows. The study, led by associate professor Maxim Topaz of Columbia University’s School of Nursing and colleagues, reviewed nearly 2.5 million biomedical papers and 97 million citations in PubMed Central. The team found more than 4,000 fabricated references across nearly 3,000 papers, with the rate rising sharply in recent years. The report cites that in 2023, one in 2,828 papers contained at least one fake reference, increasing to one in 458 by last year, and early 2026 data indicates one in 277 papers had at least one non-existent reference. Topaz described investigating a near-miss in his own work after an AI tool inserted a fabricated source. The findings reinforce the need for stronger citation verification, editorial checks, and research-assessment controls when AI assistance is used for writing and formatting.