Did Claude Discover New Biology? Anthropic’s AI Lab Claim Divides Scientists
Anthropic says 950 Claude agents found an uncatalogued DNA pattern in 21 hours, sparking disputes over novelty, credit and what counts as discovery in science.
Summary
On September 23, Anthropic said a molecular biology lab launched earlier in 2026 had made its first discovery. In 21 hours, 950 Claude agents narrowed 200,000 DNA candidates to a few and flagged a previously uncatalogued repeating pattern around a known enzyme, rather than finding a new sequence. Human scientists ran the experiments. Anthropic described the pattern as reminiscent of those behind CRISPR, but biologist Lucas Harrington argued that spotting unusual gene clusters is easier than establishing their function. Eli Lilly’s chair and CEO endorsed his criticism.
University of Copenhagen biologist Mario Rodríguez Mestre said his team had already found the pattern. Because he regularly discussed his work with Claude, Mestre questioned whether Anthropic learned from those conversations. Anthropic denied that possibility, but Mestre is discontinuing Claude use. The dispute reflects resistance to presenting AI as an autonomous discoverer rather than a tool within collaborative science. Earlier in September, OpenAI said team agents solved a million-dollar mathematics problem. Critics did not dispute the solution’s correctness, but questioned the result’s importance, while a mathematician alleged uncredited use of his work. Harrington urged companies to reserve discovery claims for fundamentally new mechanisms, even as OpenAI’s Sam Altman and Anthropic’s Dario Amodei compete to demonstrate scientific advances.
Positives
- 950 Claude agents identified an uncatalogued biological pattern after 21 hours of analysis.
- The Anthropic system reduced 200,000 DNA candidates to a few worth investigating.
- Human scientists tested the Claude agents’ hypotheses through laboratory experiments.
- A general-purpose chatbot completed candidate screening that constitutes legitimate scientific work.
- OpenAI’s team agents produced a solution to a million-dollar mathematics problem whose correctness was not disputed.
Risks & concerns
- Mario Rodríguez Mestre said his University of Copenhagen team had already discovered Anthropic’s claimed pattern.
- Lucas Harrington said identifying unusual gene clusters falls short of discovering their biological function.
- Mestre is discontinuing Claude after questioning whether his prior conversations influenced Anthropic’s result, which Anthropic denies.
- Anthropic’s comparison with CRISPR risks overstating a pattern whose mechanism and significance remain unestablished.
- A mathematician alleged OpenAI’s models used his work without credit, adding another scientific attribution dispute.