Claude Finds Novel Enzyme System with CRISPR-Like DNA Repeats in 21-Hour, 950-Agent Search
Anthropic’s life sciences research group, formed in spring 2026, has published its first major finding: Claude autonomously identified a previously uncharacterized enzyme system in bacteriophage DNA databases that shares structural hallmarks with CRISPR arrays.
The discovery centers on a reverse transcriptase (RT) gene found in a jumbo phage, already documented in prior studies, but with an associated element no one had documented: an evenly spaced array of non-coding DNA repeat sequences and an accessory protein of unknown function. The layout resembles the repeat arrays that make CRISPR programmable. Anthropic calls the new system array-associated reverse transcriptases (ART).
How the search ran
Anthropic gave Claude a high-level prompt: search a massive DNA sequence database for interesting new examples of reverse transcriptases. From there, roughly 950 Claude agents worked for 21 hours, consuming 210 million tokens. Scientists provided the initial prompt and later ran wet lab validation. Agent judgment determined which candidates were worth flagging.
One agent noticed a repeating pattern of DNA sequences sitting adjacent to an unusual RT gene. Follow-up lab testing confirmed the system as real and distinct — not a known family.
What ART might be
Reverse transcriptases copy RNA into DNA. Systems organized around programmable RT machinery have appeared before in nature, and several are now in active development as gene therapy tools that could extend CRISPR’s reach. Anthropic does not yet know ART’s primary function. A pre-print is available for independent review.
Feng Zhang, a pioneer of CRISPR genome editing at MIT and the Broad Institute: “This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation.”
Context
Anthropic’s research group includes scientists who specialize in using computational approaches to systematically read DNA, interpret its evolution, and identify biological systems for further characterization — the same methodological tradition that produced earlier CRISPR expansions and novel enzyme catalogs. Physical validation runs in the group’s wet lab facility. The model’s role is to act across the entire scientific loop: hypothesis generation, database mining, and experimental prioritization.
The ART discovery is the clearest case so far of a frontier model making an original biological finding without a human scientist directing the search at the level of individual candidates. The result arrives as Anthropic prepares for a public offering and as GPT-6 Astra’s September release has sharpened questions about which lab produces the most capable general-purpose agents. Science is now a product differentiator.