Anthropic says Claude helped identify a previously undocumented enzyme-and-DNA system after 950 AI agents searched more than 200,000 reverse transcriptases for unusual neighboring sequences. The agents ran for 21 hours, while humans chose the research question and conducted the laboratory experiments.
The candidate, named ART, consists of an enzyme near repeating DNA sequences and appears mainly in bacteriophages, the viruses that infect bacteria. Its pattern resembles systems involved in gene editing, including CRISPR, but Anthropic has not established ART’s biological function. The company presents the result as an early discovery from its Bay Area biology laboratory.
Some genome-mining researchers reject the stronger framing. Lucas Harrington, a CRISPR specialist and Mammoth Biosciences co-founder, said scientists have used similar database searches for decades and related systems have been known since 2008. In his view, finding a sequence pattern is the easier step; determining what it does is the discovery that matters.
The disagreement separates automation from scientific significance. Claude appears to have performed a large search and prioritized a candidate with limited direct guidance, which could reduce the cost of exploring biological databases. But the result remains an uncharacterized system until experiments establish its mechanism and value. Human-defined questions, laboratory validation and comparison with earlier work remain essential to judging whether the finding is genuinely new.