Study: rate of fabricated references in biomedical papers has grown 12x+ since 2023; in early 2026, one in 277 papers had at least one non-existent reference
It was a process that had become routine for Maxim Topaz. — The associate professor at Columbia University's School of Nursing …
Context & Ripple Effects
Related coverage has already documented two sides of AI’s impact on research: detectable shifts in biomedical writing and growing concern among publishers and integrity specialists that generative tools lower the cost of fabricating scientific material.
The new finding adds a concrete failure mode in the biomedical literature itself—non-existent references—moving the issue from stylistic signals and attempted fraud to errors that can enter published papers and mislead readers checking the evidence base.
First-order effects
- Biomedical authors, editors, and peer reviewers face a more immediate need to verify references rather than treating formatted citations as reliable evidence of source checking.
- Researchers who rely on affected papers may spend additional time validating underlying literature, while papers containing fabricated citations risk correction or loss of credibility.
Second-order effects
- Publishers and research-integrity teams are likely to put greater weight on citation-validation workflows, increasing scrutiny at submission and review for AI-assisted manuscripts.
- Academic users of generative AI face a sharper distinction between drafting help and evidence retrieval: unverifiable outputs create downstream work for collaborators, reviewers, and readers.
Third-order effects
- If fabricated citations continue to rise, scientific publishing may shift toward machine-checkable provenance for references and stronger accountability for authors’ verification of AI-assisted claims.
- The pattern could weaken the efficiency gains promised by AI-assisted research writing if trust costs—verification, correction, and integrity review—grow faster than editorial capacity.
The trend: Generative AI is turning research integrity from a concern about individual misconduct into an operational verification problem for the institutions that publish and use science.