Large language models give harsh and inconsistent reviews of top science papers

Major Revision, Minor Accountability: LLMs as Reviewers of Published Articles in Top-Tier Journals

Digital LibrariesArtificial Intelligence

Summary

Scientific papers are usually checked by human experts before they get published to make sure the science is good. This paper looked at how well big language models (LLMs), which can read and write text like humans, do when asked to review already published top science papers. The LLMs often suggested big changes or rejection, even though the papers had already passed expert review. They also often missed important errors that were purposely added to test them. This means LLMs currently give mixed and unreliable reviews if used to judge published science.

What this means in practice

  • For journal editorial teams: Identify limitations of LLMs when considering automated assistance in peer review to avoid relying on flawed or inconsistent feedback.
  • For scientific quality assurance teams: Use findings to guide supplemental human checks when exploring AI-assisted error detection in published papers for improving quality control.

Authors

Bastien Le Guellec, Su Hwan Kim, Raphael Bentegeac, Aghiles Hamroun

Abstract

With the growth of scientific output, the peer review system is approaching a breaking point: reviewers are increasingly hard to secure, and turnaround times lengthen accordingly. LLM-assisted or LLM-generated peer review has been proposed as a remedy, but how such models would treat manuscripts already published in top-tier journals, the closest thing the current system has to a gold standard of scientific validity, is not known. Here we assembled a corpus of 200 articles, each released under a CC BY 4.0 license and published in 2026, from 20 general-science, life-science, and clinical-medicine journals including Nature, Science, and Cell, and asked four current LLMs (two proprietary, two open-weight) to produce independent peer-review reports of each one under a blinded protocol. Their decisions were overwhelmingly harsh and starkly inconsistent with each other: 605 of 800 reviews (75.6%) recommended major revision or rejection, but the rate at which any single model did so ranged from 17.5% to 100% (p < 0.001), disagreement on manuscripts that had, by construction, already survived real editorial review at the journals in question. When the same manuscripts were doctored with one deliberately planted, field-specific methodological or statistical flaw, detection rates varied from 17.0% to 60.5% across models (p < 0.001). Recommending revision for papers that had already passed editorial review (low specificity) while missing flaws deliberately placed in front of them (low sensitivity), these models produce output that should be read for what it is: an uneven, framing- and model-dependent signal.