AI Hallucinations Have Generated Nearly 150,000 Fake Citations in Research Papers, Study Finds

What happened

A study reported by CNET has identified that AI hallucinations—fabricated outputs presented as fact by artificial intelligence systems—have generated nearly 150,000 fake citations that now appear in research papers. The discovery, characterized by scientists as a form of "AI slop" polluting academic work, indicates that fabricated references have infiltrated the formal scientific literature at a scale far beyond isolated incidents. The term "AI slop" has emerged as shorthand for low-quality, machine-generated content that degrades the information ecosystem, and its appearance in peer-reviewed contexts signals a troubling shift. Because available reporting provides only a high-level summary, critical details—including the study's authors, the methodology used to detect phantom citations, the specific journals or fields most affected, and the timeframe over which these citations accumulated—remain unclear. What is known is that the figure approaches six digits, suggesting a systemic issue rather than a handful of accidental errors.

Why it matters

Reliable citation chains are the backbone of scientific research. When a paper references work that does not exist, it corrupts literature reviews, meta-analyses, and potentially policy decisions or clinical guidelines built on that research. The apparent scale of the problem—tens of thousands of phantom citations—suggests that researchers may be using AI tools to draft literature reviews or bibliography sections without manually verifying every source, or that AI-generated content is entering the publication pipeline with inadequate human oversight. Beyond the immediate waste of time for scholars chasing non-existent papers, this erosion of trust could force journals, universities, and funding bodies to devote significant resources to citation auditing, retraction efforts, and new verification infrastructure. If readers can no longer assume that a cited paper actually exists, the entire scaffolding of evidence-based research weakens. The finding also raises questions about whether current peer-review processes are equipped to spot references that look plausible but point to nowhere.

Public reaction

No strong public signal was available at the time of reporting. Without Reddit or broader social discussion data, it is unclear whether the findings have sparked significant conversation among researchers or the general public beyond the initial news coverage.

What to watch

Observers should monitor whether major publishers and preprint servers introduce automated citation verification checkpoints or update submission guidelines to require explicit disclosures about AI use in bibliography generation. It also remains to be seen whether academic institutions will mandate training on AI hallucination risks, if existing papers will be retroactively audited, or if dedicated research-integrity startups emerge to tackle citation fraud at scale. Additionally, the response of existing academic databases—such as Crossref, PubMed, and Google Scholar—will be critical, as they may need to deploy new detection algorithms or flags for anomalous citation patterns. The next few months will likely reveal whether this study catalyzes concrete policy changes or remains a warning that the research community struggles to operationalize.

Sources

• CNET