Claude Searched 1.9 Billion Protein Clusters and Found an Enzyme System Nobody Had Described

Anthropic said on September 23 that Claude identified a previously undescribed enzyme system in the DNA of bacteriophages, the viruses that infect bacteria, per Anthropic. It calls the system array-associated reverse transcriptases, or ART: a reverse transcriptase enzyme, a partner gene beside it, and a long array of evenly spaced DNA repeats. The repeat layout resembles a CRISPR array, per TechCrunch.

The search covered about 1.9 billion protein clusters and ran 21.5 hours across 949 agent sessions, consuming 215.6 million tokens. Claude read literature and genome data, flagged the pattern, then proposed experiments to verify it. Anthropic formed its life sciences group in spring 2026, and its Bay Area lab operates at biosafety levels 1 and 2, handling no human pathogens. What ART actually does is unknown.

Reverse transcriptases copy RNA back into DNA, the trick retroviruses use and the reason the enzyme family matters in biotech. CRISPR arrays store RNA guides, which is what makes those systems programmable, so a similar repeat layout next to a reverse transcriptase is both why this is interesting and why it needs care. A structural resemblance is a hypothesis. Function comes from the bench, and Anthropic says that work is still ahead.

The method is the part other labs will copy. A 1.9 billion cluster sweep is beyond what a graduate student reads in a career, and the cost was 215.6 million tokens, a few thousand dollars at current prices. Screening at that scale turns literature search into something closer to a compute purchase.