Are fake citations really an AI problem?

 

The recent case reported by Retraction Watch involving fake references in the Journal of Academic Ethics is easy to frame as a generative AI problem. The corresponding author openly admitted that ChatGPT was used to generate the references, and nearly two-thirds of them turned out to be fabricated. That admission makes the story feel settled. AI hallucinated citations, the journal missed them, and the system failed. But that conclusion is too neat, and ultimately misleading.

What this case really shows is not that generative AI creates fake citations, but that authors chose not to verify them. Large language models do not submit manuscripts, format reference lists, or click “submit” on journal portals. Humans do. In this article, the research itself was described as real and based on real data. The failure occurred at the most basic scholarly task: checking whether cited work exists. That is not a new problem introduced by AI. It is an old one, now made more visible and easier to scale.

The details uncovered by Erja Moore make this clear. The references were not random strings. Many looked plausible. Some cited real authors writing on similar topics, but paired them with nonexistent article titles. Others listed real journals but incorrect volumes, issues, or page numbers that pointed to entirely different papers. These are the kinds of errors that slip through when references are copied uncritically, padded to satisfy reviewers, or assembled to signal familiarity with a field rather than to support an argument. Long before generative AI, scholars were already citing papers they had not read.

The uncomfortable irony is that this happened in an ethics journal, and that many fabricated references pointed to another flagship ethics outlet, the Journal of Business Ethics. But the deeper issue is not hypocrisy. It is institutional complacency. Peer reviewers typically do not check references line by line. Editors rarely audit citation accuracy unless something triggers suspicion. Publishers rely on trust, not verification, because the system was built for a slower, smaller volume of submissions. Generative AI did not break that system. It exposed how fragile it already was.

So no, generative AI is not solely responsible for fake citations. It is an accelerant, not the fire. The responsibility still lies with authors to verify sources, with editors to enforce basic standards, and with publishers to adapt integrity checks to a world where plausible text can be generated instantly. Blaming AI alone is convenient, but it lets the academic community avoid a harder conversation: that citation misconduct has always existed, and we simply no longer have the excuse of not seeing it.