New OpenAI tool renews fears that “AI slop” will overwhelm scientific research
“Science is nothing but a collective endeavor,” she said. “There needs to be some deep reckoning with what we do with a tool that benefits individuals but destroys science.”
Concerns about AI-generated scientific content are not new. In 2022, Meta pulled a demo of Galactica, a large language model designed to write scientific literature, after users discovered it could generate convincing nonsense on any topic, including a wiki entry about a fictional research paper called “The benefits of eating crushed glass.” Two years later, Tokyo-based Sakana AI announced “The AI Scientist,” an autonomous research system that critics on Hacker News dismissed as producing “garbage” papers. “As an editor of a journal, I would likely desk-reject them,” one commenter wrote at the time. “They contain very limited novel knowledge.”
The problem has only grown worse since then. In his first editorial of 2026 for Science, Editor-in-Chief H. Holden Thorp wrote that the journal is “less susceptible” to AI slop because of its size and human editorial investment, but he warned that “no system, human or artificial, can catch everything.” Science currently allows limited AI use for editing and gathering references but requires disclosure for anything beyond that and prohibits AI-generated figures.
Mandy Hill, managing director of academic publishing at Cambridge University Press & Assessment, has been even more blunt. In October 2025, she told Retraction Watch that the publishing ecosystem is under strain and called for “radical change.” She explained to the University of Cambridge publication Varsity that “too many journal articles are being published, and this is causing huge strain” and warned that AI “will exacerbate” the problem.
Accelerating science or overwhelming peer review?
OpenAI is serious about leaning on its ability to accelerate science, and the company laid out its case for AI-assisted research in a report published earlier this week. It profiles researchers who say AI models have sped up their work, including a mathematician who used GPT-5.2 to solve an open problem in optimization over three evenings and a physicist who watched the model reproduce symmetry calculations that had taken him months to derive.
