This EMBARRASSING AI-Generated Paper Exposed a Billion-Dollar Problem

A hospital librarian discovered that a published academic paper contained mostly fake citations—12 of 14 references were fabricated—highlighting how AI-assisted “hallucinated” references can slip through peer review and remain accessible long after corrections are promised. The case began when a researcher asked Jessica Wait, a hospital librarian, to locate sources listed in a single-author paper. Instead of finding the expected literature, she found that the majority of the references appeared to be non-existent, even though they looked plausible on the page.

Once the problem was identified, the paper’s lead author, Marie Atala, was contacted and initially acknowledged the issue, offering an updated reference list. But the second attempt didn’t resolve it: 16 of 20 references were still fake. A third revision eventually produced a reference list that was larger and, at least in principle, more verifiable—yet the broader system still failed. Retraction Watch and investigators then turned to Springer Nature, the publisher responsible for the paper’s online presence, only to find that the journal’s website still displayed the older, incorrect reference list rather than the corrected one.

The situation became more alarming because the references weren’t just wrong—they were effectively untraceable. When checked through tools like Google Scholar, the cited works could not be found, and even the publisher’s own site appeared inconsistent, at times returning errors such as “Sorry, we couldn’t find this article.” Despite internal claims that the matter was being handled, the paper remained live online with the incorrect citation information.

Springer Nature’s response pointed to the complexity of reference checking, arguing that authors format citations in many ways and that some automated tools can produce false positives. The underlying complaint from the investigation was simpler: if references are proven to be fabricated, retraction or correction should follow quickly and transparently. The timeline also raised eyebrows—acceptance occurred just two days before publication—suggesting there was little time for thorough verification, even as the publisher later referenced AI tools aimed at identifying irrelevant references.

The case ultimately landed on the people who rely on academic literature for real-world decisions. Jessica Wait described the impact on hospital information work and evidence-based practice: if clinicians and researchers can’t trust the citations in major publications, the downstream consequences extend far beyond academia. The story ends up as a cautionary tale about “AI-generated” shortcuts—especially when authors use large language models to draft reference lists without checking whether the cited papers actually exist—and about the gatekeeping responsibilities that journals and publishers carry once errors are detected.

In short, the scandal isn’t only that fake references were created; it’s that multiple rounds of correction still left incorrect material accessible online, forcing librarians and integrity investigators to do the verification work that should have happened before publication and should have been promptly enforced afterward.