What counts as a scientific discovery made by AI?

Meta description: Anthropic says 950 Claude agents found a DNA pattern in 21 hours, but biologists dispute whether it is a scientific breakthrough and what counts as AI discovery.
Tags: Anthropic, Claude, AI science, molecular biology, CRISPR
Featured image: Abstract molecular network visualization with DNA sequence motifs and agent-like nodes, no real people, no company logos, no text.

Last Wednesday, Anthropic announced that earlier this year it had launched a molecular biology lab. In that AI-powered lab, Anthropic Claude agents read and conjecture about hard biology problems, while human scientists run experiments based on what the agents report. The company said the lab had made its first discovery. The claim matters because it tests how much scientific credit an AI system can take before a useful lead and a claimed breakthrough are separated.

A Pattern Found, Not a New Enzyme

To understand the claim, think of a library of millions of DNA sequences, which grows as scientists sequence more of the living world. A useful lead could be a peculiar sequence that encodes an interesting enzyme; the harder task is learning what that enzyme does and how to manipulate it for something useful. After 21 hours, the company said, a system of 950 agents had not identified a brand-new sequence. It instead pointed to a repeating pattern surrounding a known enzyme, a particular pattern Anthropic said hadn't been catalogued before. Anthropic's announcement said the pattern was reminiscent of what led to the gene-editing technology CRISPR that has already transformed science and medicine, wording that makes the army of agents sound as if it found something of note.

Why Biologists Pushed Back

Some biologists were angered by that framing. One biologist's viral post, later endorsed by the chair and CEO of drugmaker Eli Lilly, said that "finding a weird cluster of genes and repeats is often the easy part. The hard part, and where the real discoveries come from, is figuring out what the system actually does." In that reading, the agents helped with some laboratory grunt work, but a discovery it is not. The episode shows that even when AI does something impressive, such as spotting a pattern in a mass of biological data that human eyes would find hard to perceive, the result may not constitute a scientific breakthrough. A finding that seems novel to an AI system can be routine, unsurprising, or simply not that consequential to a biologist.

The debate grew messier after Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said over the weekend that his team had already discovered this particular pattern, the New York Times reported. Mestre, who regularly chatted with Claude in his work, wondered whether Anthropic's team had learned from his conversations. Anthropic denies this, but Mestre says he's stopping all use of Claude anyway. The problem is partly that AI companies do not present their systems simply as tools scientists can use, like microscopes or supercomputers. They insist the AI systems are making discoveries themselves. Some researchers say that clashes with how science usually works, where new knowledge emerges from collaboration and an ever-growing arsenal of tools.

Breakthrough or Bust

That binary framing can also make people more skeptical of genuine progress when it happens. Narrowing 200,000 candidates down to a few worth exploring is no small feat; it is legitimate scientific work. It is also notable that a general-purpose chatbot could do that work, even if humans helped steer it and ultimately ran the experiments. But once the standard becomes whether Claude itself made a discovery, all of that turns into evidence in a debate with only two answers: breakthrough or bust. Once AI is judged by discovery, goalposts also shift after an apparent win.

The same pattern appeared around OpenAI. Earlier this month, OpenAI said its own team agents had cracked a million-dollar problem in mathematics. A couple of weeks later, nearly every AI skeptic in the original author's feed was sharing an article asking whether it was the math problem that really mattered. The article did not say OpenAI's solution was wrong. It argued that the particular result may not be the one mathematicians care most about. Add an accusation by a mathematician that the models may have used some of his work without credit, and people are left thinking either OpenAI cheated or the solution wasn't important anyway. Or both.

Lucas Harrington, the biologist who wrote the post critiquing Anthropic's announcement, closed with a suggestion: AI companies, he said, should "set the bar high now, so that when an AI actually discovers a fundamentally new biological mechanism, everyone appreciates how big a deal it is." Still, as OpenAI's Sam Altman and Anthropic's Dario Amodei race to one-up each other, raising the bar for scientific breakthroughs by AI might be the last thing on their minds. What to watch next is whether future lab reports distinguish a model-generated lead from an AI-made discovery. The original account appeared in The Algorithm, a weekly newsletter on AI.