AI study reveals thousands of fake references in medical research, raising patient safety concerns

A new international study using AI tools has uncovered nearly 3,000 medical research articles containing fabricated references, with experts warning the trend threatens patient safety, reports Finnish broadcaster Yle.

The research, published in the prestigious Lancet journal, scanned 2.5 million articles from early 2023 to February 2026. Among 97.1 million citations, over 4,000 were found to be non-existent in scientific databases, appearing in roughly 2,800 articles. One paper contained 30 citations, of which 18 were fake.

Laura-Maria Peltonen, associate professor at the University of Eastern Finland, stressed the findings likely underrepresent the problem: “We prioritized accuracy over comprehensiveness. The actual number of fake references in research literature is far higher.”

The study revealed a 12-fold increase in fake references since 2023, with growth accelerating in mid-2024 as AI-powered writing tools became widespread. Peltonen warned that flawed citations risk corrupting clinical guidelines, as medical recommendations rely on verifiable evidence.

“If fake references enter the evidence base, errors may unknowingly spread to patient care,” she said.

Researchers found some fabricated citations had already been referenced in other articles and appeared in systematic reviews guiding clinical treatment. They urged publishers to verify all manuscript references, retroactively screen published articles, and retract papers where fake citations undermine conclusions.

Peltonen noted that while automated verification is technically feasible, publishers currently lack incentives to implement such checks. The study also highlighted declining global trust in science, emphasizing researchers’ responsibility for verifying all content, including citations.