“I can see by eye a tandem repeat array,” a Claude agent wrote while examining DNA from a virus that infects bacteria, according to Anthropic’s September 23 account. The repeated sequence sat beside a gene for a reverse transcriptase. Anthropic called the arrangement CRISPR-like, raising a tempting question: had its AI found the makings of another gene editor?
There is a discovery here, but its boundaries matter. The reverse transcriptase itself had appeared in a 2021 peer-reviewed study. What Anthropic says Claude noticed was a nearby pattern of repeated DNA, along with a partner gene, that might mark an overlooked biological system. Neither the pattern nor the company’s early laboratory work shows that it can edit genes.
What the agents actually found
Anthropic says roughly 950 Claude agents spent about 21 hours searching DNA data, using 210 million tokens. Its preprint abstract describes a survey of reverse-transcriptase loci among 1.9 billion protein clusters. According to the company, the agents gathered more than 200,000 reverse transcriptases, selected 3,500 candidate systems and narrowed them to 20 reports for review. Those are Anthropic’s figures, not the result of an independent repeat of its search.
The candidate it named ART, for array-associated reverse transcriptases, has three reported parts in bacteriophages: a reverse-transcriptase gene, a neighboring partner gene and an array of repeated noncoding DNA. A reverse transcriptase copies RNA into DNA. CRISPR systems also contain repeated sequences, but similar-looking genomic arrangements need not do the same job. The comparison identifies a reason to investigate, not a function already established. Anthropic describes the arrangement here.
The authors report that the array produces distinct short RNAs during infection by a Staphylococcus phage. That is a physical observation beyond the original sequence search. It still does not establish what the RNAs do, whether they guide the reverse transcriptase, or whether ART can cut or rewrite DNA. Anthropic explicitly says it does not yet know the system’s function. The authors’ preprint abstract and announcement set out those limits.
The earlier human discovery
In 2021, Abby Korn, Jason Gill and colleagues described a retron-like reverse transcriptase in three jumbo phages called MarsHill, Madawaska and Machias. That work did not characterize the ART repeat array. It does mean the enzyme gene was not discovered from nothing in 2026. Anthropic’s narrower novelty claim is that an agent connected the enzyme to adjacent repeats and a partner gene that had not been recognized together as this system. The original phage study and Anthropic’s account allow those contributions to be separated.
The test behind the headline
Independent scientists interviewed by WIRED on September 29 drew the same boundary from different angles. Stanford researcher Le Cong called for more experiments. Seth Shipman said the striking part may be how the pattern was found, rather than evidence of a new CRISPR system. Gill, a coauthor of the 2021 paper, said ART had no obvious relationship to known CRISPR systems and that any gene-editing activity remained for Anthropic to prove. Those are their assessments as reported by WIRED; the Anthropic technical report had not yet been peer reviewed.
Scientists now have a specific candidate to test: does the reverse transcriptase carry out a reaction, what role do the short RNAs play during infection, and can any part of the system be programmed to alter DNA? If those experiments fail, Claude’s contribution may still be a useful demonstration of finding patterns in immense genomic databases. If they succeed, researchers will have to establish exactly what ART does before comparing its utility with CRISPR.
