Evidence that does not exist, shaping how patients are treated
In the long tradition of science as humanity's most disciplined attempt to know the world, a quiet corruption has taken root: nearly three thousand peer-reviewed medical papers have been found to cite studies that do not exist. Columbia University researchers, publishing in The Lancet in May 2026, traced this fabrication crisis to the rapid spread of AI writing tools, which can generate convincing but invented references with no mechanism for self-correction. The stakes are not merely academic — clinical guidelines built on phantom evidence shape the treatments real patients receive, and the rate of these false citations has grown twelvefold in under three years.
- An automated scan of 2.5 million biomedical papers uncovered more than 4,000 fabricated citations, with the problem accelerating sharply after AI writing tools became widely available in mid-2024.
- In one documented case, eighteen of thirty citations in a single paper were invented — and those ghost references were already being cited by other researchers, spreading through the literature like a contagion.
- Doctors and guideline developers have no reliable way to know when the evidence they trust has been fabricated, meaning patient treatment decisions may rest on studies that were never conducted.
- At the time of the audit, 98.4 percent of the affected papers had received no correction or retraction from their publishers, leaving the contaminated evidence base largely intact.
- Researchers are calling for mandatory pre-publication reference verification, retroactive screening of existing literature, and dedicated tracking categories for fabricated citations in major integrity databases.
Researchers at Columbia University have uncovered a crisis spreading quietly through medical literature: nearly 3,000 peer-reviewed papers contain citations that simply do not exist. Published in The Lancet on May 7, 2026, the study emerged from an automated audit of 2.5 million biomedical papers, which identified 4,046 fabricated references across 2,810 publications. The rate of fake citations has grown more than twelvefold since 2023, with the steepest rise beginning in mid-2024 — when AI writing tools became widely accessible to researchers.
Lead researcher Maxim Topaz, of Columbia's School of Nursing and Data Science Institute, frames the danger plainly: clinicians and guideline developers have no way of knowing when the evidence they rely on is invented. AI writing tools can generate plausible citations — complete with author names, journal titles, and publication years — that never existed. These references pass through peer review, get published, and are then cited by other researchers, eventually reaching the systematic reviews that directly shape patient care. In one examined paper, eighteen of thirty citations were fabricated, and those ghost references had already propagated into other studies.
The research team is not calling for panic, but for structural reform. They urge publishers to verify every reference before publication, create metadata systems to flag citation accuracy, and retroactively screen existing papers for corrections or retractions. At the time of the audit, 98.4 percent of affected papers had received no publisher action.
Commentators from Boston University and the University of Washington warn that the consequences reach beyond academic integrity. Public trust in science is already strained, and allowing fabricated evidence to quietly shape clinical practice deepens that wound. The audit itself — an AI system used to detect what AI tools helped create — points toward one path forward. But detection cannot undo what is already embedded in the literature, or the treatment decisions already made on evidence that was never there.
Researchers at Columbia University have uncovered a crisis quietly spreading through the medical literature: nearly 3,000 peer-reviewed papers contain citations that simply do not exist. The discovery, published in The Lancet on May 7, 2026, emerged from an automated audit that scanned 2.5 million biomedical papers published between January 2023 and February 2026. Among 97.1 million references checked, the team identified 4,046 fabricated citations scattered across 2,810 papers. The numbers alone are striking, but the trajectory is alarming. The rate of fake citations has grown more than twelvefold since 2023, with the steepest climb beginning in mid-2024—precisely when AI writing tools became widely accessible to researchers.
Maxim Topaz, an associate professor at Columbia's School of Nursing and Data Science Institute who led the study, frames the problem in terms that cut to the heart of why this matters: medical professionals and the people who develop clinical guidelines have no way of knowing the evidence they rely on is fabricated. A doctor reading a paper that cites a study proving a particular treatment works has no mechanism to verify that study actually exists. In one example the team examined, a single paper contained eighteen fabricated references out of thirty total citations. Those fake references then get cited by other researchers, propagating through the literature like a virus, eventually appearing in systematic reviews that directly inform how patients are treated.
The mechanics of how this happens are straightforward. An AI writing tool, tasked with generating a research paper on a medical topic, can produce plausible-sounding citations complete with author names, journal titles, and publication years—all invented. A busy researcher might not catch the deception. A journal's peer review process, already stretched thin, might not catch it either. The citation appears legitimate on its face. It gets published. It gets cited. It becomes part of the evidence base.
The researchers who conducted this audit are not calling for panic, but they are calling for systemic change. They recommend that publishers verify every reference submitted with a paper before publication. They want indexing services to add metadata so readers can assess citation accuracy. They urge major research integrity databases to create a dedicated category for tracking fabricated references. Most urgently, they argue that publishers should retroactively screen existing publications and issue corrections or retractions where fake citations undermine a paper's conclusions. At the time of the audit, 98.4 percent of the affected papers had received no action from their publishers.
Two senior physicians from Boston University and the University of Washington, writing in accompanying commentary, emphasize that the problem extends beyond academic housekeeping. Public trust in science is already fragile in many countries. Allowing fabricated evidence to circulate unchecked in the medical literature—evidence that influences clinical practice and patient care—threatens that trust further. They argue that authors must be held accountable for everything in their manuscripts, including the references. The responsibility cannot be outsourced to AI.
What makes this moment distinct is the speed of the problem's growth. Fake citations existed before AI writing tools became commonplace, but the twelvefold increase in less than three years suggests we are watching a new phenomenon accelerate. The tools are getting better at generating plausible text. The pressure on researchers to publish is unrelenting. The gap between what can be checked and what actually gets checked is widening. The audit itself—an AI-assisted verification system scanning millions of papers—represents one possible response: using the same technology that created the problem to detect it. But detection alone is not enough. The papers are already out there. The citations are already being used. Patients are already being treated based on guidelines built on evidence that does not exist.
Citações Notáveis
A medical professional or clinical guideline developer has no way of knowing that the evidence they are relying on does not exist.— Maxim Topaz, Columbia University School of Nursing
Authors must take responsibility and be held accountable for the entire content of a manuscript, including the references.— Howard Bauchner and Frederick P. Rivara, in commentary published with the study
A Conversa do Hearth Outra perspectiva sobre a história
Why does a fake citation matter if the paper's actual findings are sound?
Because a paper's conclusions rest on the evidence it cites. If you're claiming a treatment works and you cite five studies proving it, but three of those studies don't exist, your claim is built on air. A doctor reading that paper doesn't know which citations are real. They treat the whole thing as credible.
But wouldn't peer reviewers catch this?
In theory, yes. In practice, peer reviewers are volunteers working for free, often reviewing dozens of papers a year. They might spot an obviously wrong citation, but a plausible-sounding fake one? It's designed to pass. And if the paper's main findings seem reasonable, reviewers don't always verify every single reference.
How does this actually harm patients?
Clinical guidelines are built on systematic reviews of the literature. If those reviews include papers with fabricated citations, the guidelines are built on false evidence. A doctor following that guideline is making treatment decisions based on studies that never happened. The patient gets a treatment that might not work, or might cause harm.
Why is this happening now, specifically?
AI writing tools got very good very fast. They can generate text that sounds authoritative and includes plausible citations. A researcher under pressure to publish can use these tools to write faster. The tools don't know the difference between real and fake citations—they just generate text that fits the pattern.
What's the fix?
Publishers need to verify references before accepting papers. Databases need to flag suspicious citations. Authors need to be held responsible. But honestly, we're playing catch-up. The technology that created the problem is now being used to detect it. That's progress, but it's reactive, not preventive.