Scientists warn that AI-generated fake citations are increasingly appearing in scientific papers across global publications [1, 2].
This trend threatens the foundation of academic integrity because it introduces fabricated evidence into the peer-review process. When researchers rely on non-existent sources, it can lead to the propagation of false data and the erosion of trust in scientific discoveries.
The proliferation of these hallucinations stems from the growing use of artificial intelligence and large language models to draft research papers [1, 2]. These tools are designed to predict the next likely word in a sequence rather than retrieve factual data from a database. Consequently, the models can generate citations that look authentic, complete with plausible author names and journal titles, but do not actually exist in any library or archive [1, 2].
Academic publications are seeing these errors across various fields of study [1, 2]. The issue is compounded by the speed at which AI can produce text, allowing authors to generate vast amounts of content with references that may never have been verified by a human eye.
Experts said the risk is not limited to intentional fraud. Many researchers may unknowingly include these fake citations while using AI for brainstorming or drafting assistance [1, 2]. This creates a systemic vulnerability where the literature review section of a paper—traditionally the map of a study's intellectual heritage—becomes a collection of digital mirages.
To combat this, there are calls for more rigorous verification processes during the submission phase. This includes the use of specialized software to cross-reference citations against known databases, and a renewed emphasis on manual verification by peer reviewers [1, 2].
“AI-generated fake citations are increasingly appearing in scientific papers”
The rise of AI hallucinations in citations signals a shift in the challenges facing the peer-review system. While academic fraud has historically been a human endeavor, the automation of fabrication means that the volume of errors could outpace the ability of human reviewers to detect them. This may necessitate a transition toward 'AI-aware' auditing, where the verification of a paper's bibliography becomes as critical as the verification of its data.
