Anthropic's Claimed 'First AI Discovery' Sparks Backlash From Biologists Over What Counts as Science

What happened

Anthropic announced last week that a system of 950 Claude agents, running in a molecular biology lab the company launched earlier this year, made what it called its first scientific discovery. According to Anthropic, the agents sifted through a library of DNA sequences and, after 21 hours, flagged a repeating genetic pattern surrounding a known enzyme that Anthropic says had not previously been catalogued.

Anthropic's announcement described the pattern as "reminiscent" of the discovery path that led to CRISPR gene-editing technology. The system did not identify a new DNA sequence — it narrowed roughly 200,000 candidate sequences down to a small set worth investigating, work that human scientists then followed up on experimentally.

The pushback

The claim has drawn sharp criticism from biologists. One researcher's post, subsequently endorsed by Eli Lilly's chair and CEO, argued that "finding a weird cluster of genes and repeats is often the easy part," and that the real scientific work — determining what the pattern actually does — had not been done. Critics characterize the agents' contribution as laboratory grunt work rather than discovery.

Adding to the dispute, Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said over the weekend that his team had already identified this same pattern, according to reporting by the New York Times. Mestre, who says he regularly used Claude in his research, questioned whether Anthropic's system had learned from his own conversations with the chatbot. Anthropic denies this. Mestre says he is discontinuing use of Claude regardless.

A parallel dispute in mathematics

The biology controversy echoes a similar episode this month involving OpenAI, which said its agents had solved a million-dollar problem in mathematics. Days later, critics circulated an article questioning whether the specific problem solved was the one mathematicians actually care about — not disputing the correctness of the solution, but its significance. A separate accusation surfaced that the models may have used a mathematician's prior work without credit.

Why this matters

The underlying tension is how AI companies frame these systems: not as tools that assist scientists — comparable to a microscope or supercomputer — but as agents that make discoveries in their own right. That framing sets up a binary outcome (breakthrough or bust) that critics argue misrepresents how science normally progresses, through incremental, collaborative work using an expanding set of tools.

Lucas Harrington, the biologist whose post ignited the backlash, suggested AI companies should set a high bar now for what counts as a genuine discovery, so that when a model does eventually identify a fundamentally new biological mechanism, the claim carries weight. Given the competitive dynamic between labs like Anthropic and OpenAI, raising that bar may not be a priority for either company.

What this means

The substance of Anthropic's result appears real: narrowing 200,000 candidate sequences to a handful worth pursuing is legitimate scientific labor, and doing it with a general-purpose chatbot system in 21 hours is a meaningful demonstration of capability. The problem is the framing. By calling it a "discovery" and invoking CRISPR, Anthropic invited a fight over whether Claude itself did science — a debate that obscures the more measurable question of whether the tool sped up real research. As AI labs continue racing to claim scientific firsts, expect more of these disputes: less about whether the models did something useful, and more about whether the marketing matches the result.