Can we stop AI from flooding scientific journals? - C&EN

What the report says

Chemical & Engineering News, published by the American Chemical Society, reported July 28 that generative AI is intensifying concerns about fraudulent or low-quality scientific papers entering journals. The article, by Ananya Palivela, frames the issue as an extension of academia’s “publish or perish” incentives: AI tools can help researchers write more quickly, but they can also make it easier to fabricate manuscripts, data, citations and images at scale.

C&EN cites examples ranging from obvious AI leftovers in published papers to more polished work that may be difficult for reviewers or editors to identify. One retracted Elsevier paper reportedly began with a standard ChatGPT-style introductory phrase and still made it through peer review. The report also points to research estimating that a notable share of 2024 biomedical abstracts showed signs of AI assistance, while another study found rapid growth in papers linked to paper mills, organizations that produce fraudulent research for sale.

Experts quoted by C&EN said AI detection and screening tools can help but may not keep pace with improving models. Guillaume Cabanac of the University of Toulouse developed the Problematic Paper Screener, which looks for unusual phrases and other warning signs across large databases and has contributed to thousands of retractions, according to the article. David Resnik of the National Institute of Environmental Health Sciences said AI can be legitimate for editing or drafting with human oversight, but use should be disclosed and authors must remain accountable.

The larger question, C&EN reported, is whether policy fixes are enough. Publishers and ethics groups have adopted disclosure rules, and proposals include stronger data attestations and digital provenance for images. But the article suggests that lasting progress may require changing research evaluation systems so scientists are rewarded less for publication volume and more for the quality and reliability of their work.