Anthropic says Claude found a new CRISPR-like enzyme system

Anthropic said an autonomous search by roughly 950 instances of Claude, run over about 21 hours through a large database of DNA sequences (the preprint’s abstract puts the survey at 1.9 billion protein clusters), identified a previously uncharacterised system in bacteriophage DNA that the company said resembles the programmable architecture behind CRISPR gene editing. Anthropic named the system “array-associated reverse transcriptases,” or ART: a long array of repeating, non-coding DNA sequence sitting alongside a reverse-transcriptase enzyme and a neighbouring gene of unknown function.

The announcement doubled as an early result from a new Anthropic life-sciences research group, based in the Bay Area, in which human scientists carry out the physical wet-lab work — expressing and testing candidates — that follows from Claude’s computational search rather than replacing it. The underlying reverse transcriptase had turned up in earlier studies of a “jumbo” bacteriophage, but Anthropic said Claude “appears to be the first” to notice the accessory array and gene that give the system its CRISPR-like structure.

Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute who was not involved in the work, reviewed the finding and was quoted calling it “an exciting example of how AI agents can contribute to biological discovery,” while saying the repeat-array pattern “merits further investigation.” Anthropic was explicit that it does not yet know what the system does — whether it participates in phage defence, gene regulation or something else. Its first experiments showed the repeat array being expressed as distinct short RNAs, which it said suggested something analogous to CRISPR might be at work. The finding was published as an unreviewed preprint (later posted to bioRxiv, whose authors are all Anthropic employees and which discloses a provisional patent application), without independent replication. Kevin Blake, a Washington University School of Medicine microbiologist, was more sceptical, saying (as the Yahoo News syndication of the coverage quotes him) “There’s nothing to indicate this is a rival to CRISPR-the-technology, or could be developed into any kind of therapeutic or practical application.”

The result is one of a small number of claimed cases in which an AI system’s own search, rather than a researcher’s hypothesis, pointed to something in biology that experts had not previously described. Its significance rests on that search having found a genuine, useful pattern in public data that specialists had overlooked — a claim that stands or falls on the peer review and replication still to come, not on the announcement itself.