Last Wednesday, Anthropic said that earlier this year it had opened a molecular biology laboratory where Claude agents read about difficult biological problems and propose hypotheses while human scientists carry out experiments based on those proposals. The company also claimed that its AI-powered lab had made its first discovery.
According to Anthropic, the system involved 950 agents and produced results after about 21 hours. It did not uncover a brand-new DNA sequence; rather, the agents flagged a repeating pattern surrounding a known enzyme — a pattern Anthropic says had not previously been catalogued. The company framed the finding as “reminiscent” of the kinds of sequences that led to CRISPR, the gene-editing technology that has transformed science and medicine.
Pattern identification versus functional discovery
The announcement provoked strong reactions among biologists. In a viral post, one researcher — later backed by the chair and CEO of the drugmaker Eli Lilly — argued 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 other words: the agents did some of the laboratory grunt work, but that alone does not constitute a scientific discovery.
This episode is a reminder that an AI’s impressive technical feat — such as spotting a hard-to-see pattern in vast biological data — may not equate to a scientific breakthrough. What is novel to an AI can be routine, unsurprising, or of limited consequence from a biologist’s perspective.
Questions about provenance and credit
The story became messier when Mario Rodríguez Mestre, a biologist at the University of Copenhagen, said over the weekend that his team had already discovered this particular pattern, according to reporting in The New York Times. Mestre, who regularly conversed with Claude in his work, wondered whether Anthropic’s team had learned from his discussions; Anthropic has denied that. Mestre has said he will stop using Claude regardless.
These concerns underline a broader issue: AI companies often do not present their systems merely as tools like microscopes or supercomputers, but as entities that themselves make discoveries. That framing blurs roles and raises questions about data provenance, consent and attribution.
Recognition versus skepticism
It is also worth noting that narrowing 200,000 candidates down to a handful worth exploring is nontrivial and legitimate scientific work. That a general-purpose chatbot could perform that triage is notable, even if humans guided it and ultimately ran the experiments. But when the yardstick becomes whether Claude itself “discovered” something, the conversation polarizes into breakthrough or bust.
Similar dynamics have appeared elsewhere. Earlier this month, OpenAI said its team agents solved a million-dollar mathematics problem. Weeks later, critics were circulating articles asking whether that particular result was the one mathematicians most cared about. The critique did not assert the solution was incorrect, but questioned its significance; allegations that the models may have used some mathematicians’ work without credit further muddied public perception.
Recommendations and commercial incentives
Lucas Harrington, the biologist whose critique prompted much of the pushback, urged AI companies to set a high bar now so that when an AI truly discovers a fundamentally new biological mechanism, everyone will understand the magnitude of the achievement. Yet the commercial and reputational race among firms — exemplified by figures such as OpenAI CEO Sam Altman and Anthropic co-founder Dario Amodei — may incentivize quick, attention-grabbing announcements over careful, conservative standards.
Conclusion
The Anthropic Claude episode highlights that assessing AI-generated results is not purely technical: it involves questions of scientific method, attribution, and communication. Identifying patterns and understanding their biological function are distinct steps, and many scientists want those steps to be clearly separated so the community can reliably judge when AI has truly delivered a breakthrough.
