Anthropic said on Sept. 23, 2026, that AI agents built on its Claude model flagged a previously uncharacterized enzyme system it calls array-associated reverse transcriptases, or ART. The agents, which carry out tasks on their own, were working with the company's new life-sciences research group. A leading AI lab is saying its AI agents helped spot a genome pattern that looks CRISPR-like, though the company itself says ART's function is still unknown and the underlying enzyme had been identified before. Anthropic published the claim on its news page with a preprint, a paper not yet peer reviewed, hosted on its own site.
As Anthropic describes it, ART is a reverse transcriptase, a kind of enzyme, found mainly in bacteriophages, viruses that infect bacteria. It is paired with a partner gene and a long array of evenly spaced DNA repeats that resemble a CRISPR array. "CRISPR-like" describes that layout of repeats; it is not a claim that ART is a proven new gene-editing tool. Anthropic says the jumbo-phage reverse transcriptase itself had been identified before, and what Claude is credited with noticing is the repeat array and the partner gene. Early lab work, the company says, shows the array is expressed as distinct short RNAs, which is suggestive but does not yet explain how ART works.
Anthropic says it gave Claude a broad prompt to search DNA databases for interesting reverse transcriptases, then let the agents investigate families of them and write up candidates. One search ran about 21 hours across roughly 950 agents before an agent flagged a pattern of repeats beside an unusual reverse transcriptase. That search used about 210 million tokens, the small chunks of text, often parts of words, that AI models read and write.
People do the hands-on lab work in a Bay Area lab limited to Biosafety Level 1 (BSL-1) and Biosafety Level 2 (BSL-2) work, and the team does not handle pathogens that infect humans, Anthropic says. Feng Zhang, quoted in Anthropic's announcement, called the RNA-repeat arrays found with reverse transcriptases genuinely intriguing and worth further investigation.
The preprint has not been peer reviewed, no outside group has yet repeated the genome search, and there is no lab proof of what ART actually does. For now, Anthropic's account, Zhang's encouragement and the CRISPR comparisons in news coverage do not show that a new gene-editing tool has arrived.