One of Claude's agents was combing through a database of bacteriophage DNA when it flagged a suspiciously tidy repeat sequence and noted, in its own words, that it looked like "a CRISPR-like... repeat array" sitting next to an unusual reverse transcriptase gene. Weeks of lab work later, that log entry turned into an official announcement from Anthropic.
The finding comes out of a life-sciences lab the company quietly opened in the Bay Area this past spring. It runs at BSL-1 and BSL-2 — no human pathogens involved — and pairs Claude Science and Claude Code with human researchers: the AI hunts through DNA datasets for uncharacterized protein families and proposes hypotheses, while people run the actual wet-lab experiments.
This particular discovery took roughly 950 agents working in parallel, 21 hours, and 210 million tokens. They were scanning variants of reverse transcriptase, an enzyme that copies RNA back into DNA, when they landed on an unusual version from a large "jumbo" phage. Next to that gene sat an evenly spaced array of DNA repeats, a partner gene, and short RNAs transcribed from the array itself. Anthropic named the system array-associated reverse transcriptases, or ART.
The layout does echo CRISPR's signature setup — a repeat array paired with an enzyme that uses RNA as a guide. That doesn't mean ART can cut genes the way Cas9 does; what it actually does inside the cell is still unknown, and the lab's experiments are ongoing.
Outside reaction has stayed measured. Feng Zhang, the MIT and Broad Institute scientist who helped pioneer CRISPR gene editing, called the discovery "intriguing" and said it "merits further investigation" — polite scientific language for "not proven yet." Anthropic CEO Dario Amodei was careful to note that Stanford researchers had already described a similar system, so Claude's agents effectively confirmed a parallel line of research rather than starting from zero. He also stressed that humans still run every physical experiment; the AI isn't controlling lab equipment on its own.
If ART's function checks out, biologists could gain another genetic tool alongside CRISPR rather than a replacement for it. For now, it's a hypothesis waiting on bench-top proof, not a finished technology.
