On September 23, 2026, Anthropic announced a new life sciences research group and laboratory, and with it a preprint, “Autonomous AI agents discover reverse transcriptases with tandem repeat arrays” (Peter H. Yoon, Januka S. Athukoralage, Emmanuel Ameisen, Eric Kauderer-Abrams, Nicholas T. Perry and Matthew G. Durrant, all at Anthropic). The team deployed Claude Code instances in an agentic harness to survey reverse transcriptase (RT) loci across 1.9 billion metagenomic protein clusters. Reverse transcriptases copy RNA into DNA, and bacterial RT systems have previously yielded biotechnology tools, so the aim was genome mining: collect, classify and investigate sequences at a scale that manual expert curation cannot reach.
By Anthropic’s account, roughly 950 agents spent 21 hours and 210 million tokens on the search. They gathered more than 200,000 RTs, picked out 3,500 new candidate systems and narrowed those to 20 candidates written up as human-readable reports. Among the top candidates was a family the authors call array-associated RTs, or ART: jumbo-phage RTs coupled with a dedicated partner gene and an array of roughly 200-nucleotide repeat units reminiscent of CRISPR arrays. The preprint says session transcripts show an agent found the family by reading the DNA beside a phage RT and noticing an unannotated tandem-repeat array, a feature no predefined pipeline was looking for, and attributes the behavior to specific Mythos 5 internal signals that respond to repeated DNA.
Human scientists then took the candidate to the bench, expressing proteins in standard laboratory strains and characterizing them biochemically and structurally. The reported wet-lab result is that ART arrays are highly expressed and appear as discrete short RNA units during Staphylococcus phage infection, suggesting an RT system directed by a repertoire of distinct RNAs. Feng Zhang of MIT and the Broad Institute, after reviewing the preprint, called it “an exciting example of how AI agents can contribute to biological discovery.”
The significance is where the agent sat in the process: not executing a fixed pipeline faster, but doing the anomaly-spotting step that genome mining has always left to expert curators. What the work does not show is what ART does. Anthropic states plainly that it does not yet know the system’s function, the lab work was done by humans, and this is a lab preprint about its own model rather than a peer-reviewed paper. Discovering an unusual gene arrangement is the start of a biology story, not the end of one.