Claude agents spot CRISPR-like repeats beside a phage enzyme

Anthropic said on September 23, 2026, that a team of Claude agents had found an enzyme system in bacteriophage DNA that no one had described. Its announcement post says about 950 agents searched for 21 hours. One of them spotted a repeating DNA pattern that resembles a CRISPR array.

It's the first result from Anthropic's biology research, Reuters reported. It came days after the news that the company had quietly opened a wet lab in the Bay Area.

The agents found the repeats by reading raw DNA

The system is built around a reverse transcriptase, an enzyme that copies RNA into DNA. That enzyme, found in large viruses called jumbo phages, was already known. The agents noticed what sits beside it: an array of evenly spaced repeats and a partner gene of unknown function.

Anthropic calls it array-associated reverse transcriptases, or ART. According to the research pre-print, scientists wrote one research brief, and the agents ran on their own from there, using 215.6 million tokens.

The agents sorted about 200,000 reverse transcriptase clusters and ended with 19 reports for human review. In one, an agent read the DNA next to an odd enzyme and wrote that it could see a repeat array by eye.

The pre-print shows the find was hard to repeat

The pre-print adds details the post leaves out: Anthropic ran the same search ten more times, and every rerun missed the array. In separate tests, the strongest Claude models spotted it mainly when they read long stretches of raw DNA.

The paper also says no CRISPR-associated genes sit near any ART locus. The team hasn't shown that the enzyme is active, and the system's biological function is still unknown.

Outside scientists credit the method more than the molecule

Feng Zhang of MIT and the Broad Institute praised the work in a statement Anthropic published, but others were more cautious. Stanford's Le Cong told Wired's Emily Mullin: "The experiments are still in the queue. The PR is already live."

Seth Shipman of the Gladstone Institutes told Wired that "the novel thing is how they found it, not what it is." Jason Gill of Texas A&M, whose team reported the enzyme in 2021, said ART appears to have no obvious link to known CRISPR systems.

The Hacker News thread drew 805 comments. Of the top 30 on October 11, about a third questioned whether Claude did it alone or whether the system is new. About a quarter called the post marketing or thin on rigor.

A sober framing would be something like: Claude identified a previously undescribed genomic arrangement around a known reverse transcriptase.

Spacecosmonaut on Hacker News, September 23

That's an early snapshot of developer reactions, not a survey.

Anthropic's Bay Area lab, which works only at biosafety levels 1 and 2, is running the experiments to learn what ART does. The team works in Claude Science and Claude Code, and it's asking outside scientists for research proposals.