Anthropic's AI biology lab claims its first discovery and biologists push back

Anthropic's lab and its claimed first find

Anthropic has built a molecular biology lab where Claude agents pore over hard problems in biology while human scientists run the experiments the agents propose. As MIT Technology Review reports, the company announced that this setup had produced its first discovery.

The specifics matter. According to MIT Technology Review, a system of 950 agents worked for 21 hours and surfaced a repeating pattern around a known enzyme. It was not a previously unseen sequence; the agents flagged structure surrounding something already on the books, though Anthropic said that particular pattern had not been catalogued before. The company's announcement drew a parallel to the observations that led to CRISPR, the gene-editing technology that reshaped medicine, which made the result sound far more consequential than the underlying finding described.

Biologists push back

The framing irritated researchers. MIT Technology Review highlights a viral post by biologist Lucas Harrington, later endorsed by the chair and CEO of the drugmaker Eli Lilly, arguing that spotting a strange cluster of genes and repeats is frequently the easy part of the job. The difficult, discovery-defining work comes afterward, in determining what the system actually does. By that standard, the agents handled a useful piece of screening rather than achieving a breakthrough.

The piece's wider point is that novelty for a machine is not the same as novelty for a field. An AI system may genuinely identify a pattern in vast biological data that no human would spot unaided, and the result can still be routine, unsurprising or simply unimportant to working biologists.

A dispute over priority

The claim has also drawn a direct challenge. According to the New York Times, as cited by MIT Technology Review, Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said over the weekend that his team had already found this same pattern. Mestre, who regularly used Claude in his own research, said he wondered whether Anthropic's team had absorbed the insight from those conversations. Anthropic denies it, but Mestre says he has stopped using Claude entirely.

Tools or discoverers

MIT Technology Review locates the deeper friction in how AI companies present their systems. They are not marketing them as instruments, comparable to microscopes or supercomputers, but insisting the systems make discoveries on their own. That stance sits uneasily with how science typically advances, through collaboration and an expanding set of tools rather than lone acts of discovery.

The framing also sells short work that deserves credit, the piece argues. Narrowing some 200,000 candidates down to a handful worth pursuing is real scientific labor, and the fact that a general-purpose chatbot did it is notable even with human steering and humans running the experiments. But once the yardstick becomes whether Claude itself discovered something, the nuance collapses into a binary verdict: triumph or nothing.

The same dynamic plays out after apparent wins. Earlier in the month, OpenAI said its agents had solved a million-dollar mathematics problem. Weeks later, MIT Technology Review notes, skeptics circulated an article questioning whether that particular result was the one mathematicians actually cared about. That piece did not claim the solution was wrong. Combined with a mathematician's accusation that the models drew on his work without credit, readers were left concluding either that OpenAI cheated or that the result did not matter. Possibly both.

Why it matters

This episode is really about the terms on which AI-assisted science will be judged. Overclaiming invites backlash from domain experts and erodes trust, and it makes the public more skeptical even when genuine progress arrives. Under-crediting, meanwhile, obscures real capability gains in tasks like large-scale candidate screening. Harrington's suggested fix is for AI companies to set a high bar now, so that when an AI does uncover a fundamentally new biological mechanism, everyone recognises the magnitude. As MIT Technology Review dryly concludes, with Sam Altman and Dario Amodei racing to outdo each other, that may be the last thing on their minds.