AI agents are checking the scientific literature — and spotting decades-old errors
https://www.nature.com/articles/d41586-026-02235-8
AI is taking on a new role in science: checking the science itself. Researchers are using AI agents to scan papers, databases and experiments for errors that humans may have missed, including decades-old mistakes that have become part of the scientific record. In one case, an AI system flagged boiling-point values that researchers later confirmed were wrong.
The potential is huge. AI can examine scientific literature at a scale humans simply can’t match, helping researchers catch errors before they spread into future work. But there’s a catch: AI checkers make mistakes too. They can flag correct findings, miss genuine errors and even invent problems that aren’t there. For now, researchers see AI as a useful second set of eyes—not a replacement for human judgement. The bigger question may be how we build systems that can help us check what we think we know without creating new errors along the way.
