AI-Generated Fake Citations Surge in Academic Papers, Raising Questions on Consequences

Somewhere in a university office in Lima, a scientist received a letter that should never have arrived. A researcher whose name appeared in the bibliography of a published article wrote to say the work cited didn't belong to him—it was invented. The article, published in a journal indexed by Scopus, would soon be retracted. This wasn't an isolated mistake. It was a symptom of something spreading through the global scientific record like a quiet infection.

Artificial intelligence doesn't yet know how to create new scientific knowledge, but it has become remarkably skilled at fabricating the appearance of it. The systems that power modern AI can hallucinate—generate plausible-sounding but entirely false information—with such confidence that they fool both machines and people. For a casual user, this is an annoyance. For science, it's a crisis.

The numbers tell the story. In 2023, roughly one in every 2,828 academic articles contained at least one fabricated citation. By 2025, that ratio had collapsed to one in 458. In the first weeks of 2026, it had worsened further: one in 277 articles—or about 57 per 10,000 publications. That's a more than twelvefold increase in just three years. A study published in The Lancet documented the surge. Nature magazine, using advanced detection software from a company called Grounded AI, analyzed 4,000 articles from five major scientific publishers. When researchers manually reviewed the 100 most suspicious papers, they confirmed that 65 contained at least one invalid reference. If that proportion holds across all academic publishing, more than 110,000 of the roughly 7 million papers published in 2025 likely contain fabricated citations.

Peru is not exempt. Percy Mayta, the scientific editor and vice rector of research at Universidad Científica del Sur, has watched this unfold firsthand. His journal, Desde el Sur, caught a Chilean article where one of the supposedly cited authors contacted them directly: the work attributed to him didn't exist. It was pure invention. The article is being retracted. Mayta explains that scientific fraud has always existed, but AI has accelerated it dramatically. Before, dishonest researchers had to work harder to hide their deceptions. Now, they can generate entire bibliographies in seconds.

The problem runs deeper than simple laziness or corner-cutting. AI hallucinations follow patterns. Researchers once spotted fabricated work through small stylistic oddities or awkward phrasing. Now the deception is systematic. Early warning signs included what academics call "tortuous phrases"—clumsy synonyms and meaningless jargon that AI used to disguise its patterns. Then came "humanizer" tools designed to rewrite AI text so it reads as if written by a person. But the real problem has moved beyond writing style entirely. The smoking gun is now the citation itself: references that don't exist, papers that were never published, authors who never wrote what they're credited with.

Detecting this fraud is harder than it should be. Standard plagiarism detection tools like Turnitin aren't designed to catch AI-generated text, let alone fabricated references. Universities are developing policies faster than they can implement them. There's a widening gap between how quickly students adopt AI and how quickly their teachers understand it. At Científica del Sur, fewer than 30 percent of researchers were using AI or even familiar with it a year ago. That's changing, but slowly. The real solution, Mayta argues, requires expertise. You can't spot an AI hallucination if you don't understand the subject deeply enough to know what's plausible and what isn't. For now, the work remains manual: advisors, editors, and researchers must still read through bibliographies and verify that each citation actually exists and actually says what the author claims it says. It's tedious. It's time-consuming. But there's no alternative yet.

Mayta is clear about what AI should and shouldn't do in academic work. The tool can help search and organize information, check translations, review writing for coherence, even verify that citations are formatted correctly. What it shouldn't do is invent the research itself. When researchers ask AI to do the heavy lifting—to generate arguments, fill in gaps, create evidence—that's when hallucinations become fraud. A fabricated citation isn't a mistake. No one invents a reference by accident. It's an act of deception, and it should carry consequences. The real danger isn't the technology itself. It's that the speed of AI development now outpaces our ability to detect and regulate it. The scientific community's greatest strength has always been trust—trust that results are real, that methods are sound, that the evidence actually exists. Fabricated citations destroy that. They're not just bad papers. They're a threat to the foundation of how science works.