AI 'Slop' Infiltrates Academic Research as Peer Review Fails to Filter Fabrications
Researchers find a majority of reviewed AI conference papers contain fake citations and authors, yet some still secure top honors.
The integrity of the scientific record is facing a systemic crisis as large language models (LLMs) are increasingly used to automate both the creation and evaluation of academic research. This "slop" cycle—where AI-generated content is submitted to be reviewed by AI-assisted peers—is allowing fabricated data and fake identities to bypass traditional quality controls.
Researchers Caleb Robinson and Isaac Corley recently reported that 68% of the 22 AI conference papers they reviewed across NeurIPS, WACV, and TerraBytes contained fabricated citations, fake author lists, or clear LLM-generated nonsense. These findings suggest a critical breakdown in the vetting process; specifically, two papers flagged by the researchers for having fake authors were still accepted as oral presentations, one of the highest honors at these venues.
The Automation Loop
This surge in academic fabrication coincides with a massive increase in submission volumes that has made human oversight nearly impossible. To manage the load, the industry is leaning further into the very technology causing the problem. For instance, NeurIPS 2026 is currently conducting an AI-assisted review experiment that allows reviewers to use LLMs integrated directly into the OpenReview platform.
This creates a precarious feedback loop: authors use LLMs to generate plausible-sounding but fake research, and reviewers use LLMs to summarize or write the evaluations. When the human element is removed from both ends of the process, the system loses its ability to distinguish between a breakthrough and a hallucination. This erosion of rigor transforms the peer-review process from a filter for truth into a mirror for AI-generated noise.
Industry Implications
The acceptance of fraudulent papers as oral presentations suggests that the peer-review system is no longer effectively filtering for truth or validity. If the premier conferences in the field cannot identify fake authors or fabricated citations, the risk of AI-generated misinformation becoming embedded in scientific literature increases. This threatens to pollute the foundation of future research, as subsequent scientists may build upon "findings" that never existed, leading to a cascade of errors across the global research community.
What's Next
As the crisis deepens, researchers are turning to automation to fight automation. Isaac Corley has developed "bib-audit," a tool designed to automatically detect fabricated references in academic papers. The industry must now determine if these automated audits can scale fast enough to keep pace with the volume of AI-generated submissions, or if the academic publication loop has already been compromised beyond repair. The outcome will determine whether the scientific method can survive the era of generative automation.