Anthropic Says Its AI Made a Biology Discovery — Scientists Push Back

Anthropic, the company behind the Claude chatbot, said it set up a molecular biology lab where 950 AI "agents" (AI that can take actions on its own) spent 21 hours reading through DNA sequences. They flagged a repeating pattern around a known enzyme that the company said had never been catalogued before. Anthropic compared the finding to the early work behind CRISPR, the gene-editing tool that reshaped medicine.

Many biologists were not impressed. A viral post from biologist Lucas Harrington argued that "finding a weird cluster of genes and repeats is often the easy part." The hard part, he wrote, is figuring out what that system actually does — and that's where real discoveries come from. The chairman and CEO of drugmaker Eli Lilly endorsed that criticism. Meanwhile, a University of Copenhagen biologist, Mario Rodríguez Mestre, told the New York Times his own team had already found this pattern, and he wondered whether Anthropic's team picked it up from his past chats with Claude. He's now quitting the tool. Anthropic denies learning from his conversations.

The deeper problem is how AI companies frame these results. Instead of presenting AI as a tool — like a microscope or a supercomputer — they insist the AI itself is making discoveries. That sets up a two-answer debate: breakthrough or bust. Earlier this month, OpenAI said its agents solved a million-dollar math problem, and skeptics quickly argued the solution may not be the result mathematicians actually cared about.

Here's the fair way to see it: whittling 200,000 DNA candidates down to a few worth testing is genuine, useful scientific work — and it's notable that a general-purpose chatbot could help do it. But calling it a "discovery" overstates things, and it makes people cynical about real AI progress when it finally arrives.