Academic publishing is experiencing a significant increase in AI-generated content, raising concerns about the integrity of the scientific record. Estimates suggest that 57 percent of published academic articles in 2025 exhibited detectable evidence of large language model influence, marking a substantial rise from 12 percent in 2023. This surge is overwhelming the traditional peer-review process, which is managed primarily by volunteers.
To test the current defenses of publishers, a writer created two completely fabricated scientific papers and submitted them to the journal IEEE Access. The first paper was generated using free chatbots with simple prompts. Although the main text appeared highly convincing to reviewers, the journal's automated integrity checks quickly rejected the submission because the artificial intelligence had hallucinated numerous academic references.
A second paper was generated using an autonomous research system designed to conduct computational experiments and write the results without human intervention. While this advanced system did not fabricate citations, the journal rejected the paper due to its heavy reliance on non-peer-reviewed preprints and subtle visual artifacts in the generated figures.
Experts caution that as artificial intelligence tools become more sophisticated, distinguishing genuine scientific discovery from synthetic outputs will become increasingly difficult. Consequently, the scientific community may need to adopt new verification methods that authenticate the actual laboratory work and raw data, rather than relying solely on post-publication peer review.