Anthropic says Claude has found a previously undescribed enzyme system in the DNA of bacteriophages, the viruses that infect bacteria. Next to the enzyme’s gene sits a long run of repeating DNA that resembles a CRISPR array. The company does not yet know what the system does.

Anthropic announced the result on Wednesday in a blog post and a preprint that has not been peer reviewed. It is the first result from a new molecular biology group and lab inside the company. The team calls the system array-associated reverse transcriptases, or ART.

A reverse transcriptase is an enzyme that copies RNA into DNA. Bacteria often use these enzymes to defend against viruses. ART has three parts: the enzyme, a partner gene beside it, and an array of evenly spaced DNA repeats.

The layout recalls CRISPR, where an array holds a bank of different RNA guides. According to the preprint, ART arrays hold 3 to 21 copies of a short repeat. None has the cas genes that CRISPR systems carry nearby.

In lab tests, Anthropic’s scientists found that the array is read out as a set of distinct short RNAs. In published data from a phage, these RNAs made up as much as 8% of the phage’s RNA 15 minutes after infection. The preprint says the team has not yet shown that the enzyme is active, or that it works on these RNAs.

Anthropic says only a handful of known systems share this set of features. All of them can cut, copy or paste DNA.

How the search ran

The team gave Claude one research brief: find new reverse transcriptase systems in a database of 1.9 billion protein clusters. Agents running Claude Mythos 5 did the rest. One agent planned and ran each task while a second reviewed it, and agents opened new tasks as they went.

The campaign ran for 21.5 hours without human input, across 949 agent sessions and 215.6 million tokens, the preprint says. The agents recovered about 200,000 enzyme clusters, scored 3,564 candidate partner families and filed 19 reports for human review.

ART came from a side path. An agent reading raw DNA beside an unusual enzyme wrote that it could see a repeat pattern “by eye”. It then counted the repeats, compared them with known systems and searched the literature before filing its report.

Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, reviewed the preprint. Anthropic published his comment with its post.

“The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation,” Zhang said.

What is not yet known

The enzyme itself was not new. Earlier studies had identified it, Anthropic says, and Claude appears to have been the first to spot the array and the partner protein. The preprint notes that the first genome report on one ART phage described the enzyme but not the repeats or the partner gene.

The find was also hard to repeat. Anthropic ran the same campaign ten more times. No rerun read the DNA upstream of the enzyme, and all ten missed the array. The authors put this down to the size of the search and the agents’ unpredictable behaviour.

In fixed tests, Anthropic’s four most capable models described the array in at least 90% of attempts when given the DNA directly. With files and tools, the rate fell as low as 32%. Often the model never read enough DNA to see a full repeat.

The team also looked inside Mythos 5 as it read the sequence. Two internal signals fired on the repeats just before the agent named them. Anthropic has built tools before to study Claude’s unspoken thoughts.

The lab behind it

Anthropic formed the group in spring 2026. Its Bay Area lab works only at biosafety levels 1 and 2 and handles no pathogens that infect humans, the company says. Human scientists do all the lab work. The team uses Claude Code and Claude Science, which Anthropic launched in June.

The group sits in Anthropic’s life sciences unit, which also works on drug discovery. Anthropic paid $400 million for Coefficient Bio in April. OpenAI launched GPT-Rosalind, a model for life sciences research, the same month.

Anthropic says more experiments are under way to learn what ART does. It is asking other scientists to send research proposals.