Is Your Paper Being Reviewed by an LLM? Benchmarking AI Text Detection in Peer Review

Researchers specializing in academic integrity published a study on the arXiv preprint server in February 2025 addressing the growing concern that human reviewers are increasingly using Large Language Models (LLMs) to generate peer reviews for scientific manuscripts. The report warns that the fundamental trust in the academic publishing system is under threat as experts potentially outsource their critical evaluations to AI tools rather than providing original human oversight. This investigation was prompted by the rapid advancement of generative AI, which has made it easier for negligent reviewers to bypass the rigorous intellectual effort traditionally required for high-stakes scientific validation. The study highlights that peer review serves as the primary barrier against the publication of flawed or fraudulent research, a process built on the assumption that subject-matter experts provide nuanced analysis. If LLMs are used to simulate this process, the results may lack the deep technical scrutiny necessary to verify complex methodology or identify subtle errors. The researchers set out to benchmark the effectiveness of AI text detection tools specifically within the context of scholarly review, seeking to determine if journals can reliably identify whether a critique was written by a person or a machine. Furthermore, the paper explores the broader implications of this technological shift for the scientific community, noting that the automation of feedback could lead to a homogenization of scientific thought. While AI can assist with grammar or formatting, its use in evaluating the actual merit of a study poses a risk to the diversity and depth of academic discourse. The findings suggest that as LLMs become more sophisticated, the scientific community may need to implement stricter metadata monitoring and more robust verification protocols to preserve the transparency and accountability of the peer review era.