When can AI be said to make a scientific discovery?

Anthropic recently said its Claude agents helped uncover a previously uncatalogued pattern around a known enzyme in a molecular biology project. The company framed the result as an AIassisted discovery, but some biologists argued the work was more about sorting data and proposing leads than making a true breakthrough. The debate highlights a broader question: when does AI cross the line from being a useful research tool to being the thing that actually discovers something new? In this case, critics said the hard part of science was still understanding what the pattern means and proving its significance in the lab. The discussion has also been complicated by claims from other researchers that the pattern may have already been found independently. More broadly, the article argues that AI companies may be setting a low bar by treating pattern detection or problem narrowing as scientific discovery, even when human scientists still guide the work and test the results.