Anthropic says Claude agents found a biological pattern; biologists dispute the claim

Last week Anthropic announced it had set up a molecular-biology laboratory staffed by 950 Claude agents that read scientific literature, generate hypotheses, and hand them off to human scientists to run the experiments. According to the company, the system spotted a recurring pattern around a known enzyme within 21 hours — a pattern Anthropic says had never been catalogued. The firm compared the finding to the early steps that eventually produced CRISPR gene-editing technology, which has already reshaped science and medicine.

Biologists did not accept the framing. A viral post by one researcher, endorsed by Eli Lilly chairman and chief executive David Ricks, argued that spotting unusual gene clusters is the easy part. The hard part — the part where real discoveries are born — is figuring out what the system actually does. The agents may have saved preliminary screening work, critics said, but that does not make it a discovery.

The story grew more complicated when Mario Rodríguez Mestre, a biologist at the University of Copenhagen, told the New York Times that his group had already found the same pattern. Rodríguez Mestre, who regularly chatted with Claude as part of his work, wondered whether Anthropic's team had learned from those conversations. The company denies it, but Rodríguez Mestre said he would stop using the model.

The core problem is conceptual, not technical. AI companies insist on presenting their systems as entities that "discover" on their own, rather than as tools — like microscopes or supercomputers — that scientists operate. That stance clashes with the way science actually advances: through collaboration, the accumulation of knowledge, and a toolkit that grows over time. When the only metric is "did the AI discover," every intermediate achievement becomes ammunition in a binary argument: breakthrough or failure.

That dynamic erodes the ability to assess real progress. Winnowing 200 thousand candidates down to a few dozen worth of deep investigation is a legitimate scientific achievement, even if humans guided the process and ran the experiments. The fact that a general-purpose chatbot can perform that work is noteworthy. But the moment the question is phrased as "did Claude itself discover," the achievement gets swallowed by a symbolic debate.

A parallel example arrived earlier this month in mathematics. OpenAI claimed its agents had solved a million-dollar problem. Two weeks later, nearly every AI skeptic in the feed was sharing an article asking whether that particular problem mattered at all. The article did not argue the solution was wrong, only that the result might not be the one mathematicians had been waiting for. When the standard is "AI discovery," even a technical victory turns into a philosophical debate instead of a progress metric.