Anthropic's Claude autonomously flags a new CRISPR-like enzyme system in bacteriophages

Anthropic has stood up a life-sciences research group and a wet lab in the Bay Area, and its first notable result is a genomics find made largely by AI. Given only a high-level prompt to comb a massive DNA sequence database for interesting reverse transcriptases, roughly 950 Claude agents spent 21 hours and 210 million tokens winnowing over 200,000 candidates down to about 3,500 new systems and then to 20 worth deep analysis. One agent flagged an odd reverse transcriptase from a jumbo phage sitting next to a repeating array of non-coding DNA plus an accessory protein of unknown function—a combination the team calls array-associated reverse transcriptases (ARTs).

The significance is in the pattern, not yet the function. The RT itself was already known, but Claude appears to be the first to spot its defining architecture, which structurally echoes CRISPR: an enzyme paired with a repeat array. That configuration has only ever shown up together in a small number of systems, all of which are programmable and can cut, copy, or paste DNA—the same class of machinery that gave biotech restriction enzymes, PCR, and gene editing. CRISPR pioneer Feng Zhang called the RNA-repeat-array finding genuinely intriguing and worth further study, while stopping short of any functional claim.

The broader story is about workflow. Anthropic is positioning agents as collaborators at every step of genome mining—reading literature, reproducing known results to validate their methods, generating and then critically pruning hypotheses, and drafting human-readable candidate reports, work that normally takes an expert weeks to months. Humans still run all the bench experiments, and the lab restricts itself to BSL-1/BSL-2 work with no human pathogens. The team has posted a pre-print, and cautions that ART’s actual biological role remains unknown.