AI beats scientists at finding errors in clinical research papers

A study posted as a preprint on medRxiv found that large language models outperformed human reviewers at identifying deliberately inserted methodological errors in clinical research papers. In the test, Anthropic's Claude Opus 5.5 spotted 96 percent of planted errors, while human reviewers found only 27 percent. Researchers and experts suggest that AI checks could speed up journal evaluations and improve the quality of published studies, though realizing this potential may require changes to medical journal policies that currently ban human reviewers from using AI. Open article →