AI language models are generating fake citations that infiltrate peer-reviewed scientific literature, creating a credibility crisis in academic publishing. Researchers warn that citations appearing in published papers increasingly lead to nonexistent studies, fabricated sources, and invented authors.
Large language models like GPT-4 and similar systems produce plausible-sounding references as part of their text generation process. When researchers use these tools to draft papers without verification, hallucinated citations slip past peer review and into published work. The problem spans multiple disciplines and journals, contaminating the foundation of scientific knowledge that citations represent.
The mechanism is straightforward. Language models predict likely text sequences based on training data. When asked to generate citations, they produce formatted references that sound authentic but lack verification against actual databases. Authors relying on AI writing assistance often fail to check citations manually, assuming the model's output is accurate. Peer reviewers, overwhelmed with submissions, cannot feasibly verify every reference.
This undermines the core function of citations. They serve as traceable links to prior research, allowing readers to verify claims and build understanding on established foundations. Fake citations break this chain. When future researchers cite papers containing fabricated references, they propagate misinformation without knowing it.
The problem affects journals across physics, chemistry, medicine, and computer science. Some publishers report citation error rates rising sharply in the past two years, correlating with increased AI adoption in manuscript preparation. A few journals have begun implementing AI detection tools, but these catch only obvious cases.
Solutions remain limited. No universal verification system exists that scales across millions of citations. Training researchers to manually verify AI-generated citations helps but proves tedious and time-consuming. Some propose requiring disclosure when AI tools assisted in writing. Others advocate stricter peer review protocols specifically checking citations.
The crisis highlights a fundamental tension. AI writing tools improve efficiency and accessibility for researchers worldwide. Yet deployed without safeguards,