Give a swarm of AI agents a high-level homework assignment — dig through DNA for weird reverse transcriptases — and one of them may yell (in text) that the sequence next door looks spectacular. That is roughly how Anthropic’s new life-sciences lab says Claude found ART: array-associated reverse transcriptases, a previously uncharacterized system in bacteriophages that looks a little like CRISPR’s famous repeat arrays.
Over about 21 hours, roughly 950 agents burned through some 210 million tokens, surveying hundreds of thousands of reverse transcriptases and narrowing candidates until one agent spotted tandem DNA repeats beside an odd-looking RT gene, plus a partner gene and short RNAs. The underlying RT had been seen before; Claude appears to be the first to notice the full package of defining features.
CRISPR pioneer Feng Zhang called the RNA-repeat find “genuinely intriguing” and said it merits further investigation — praise with a scientist’s built-in pause button.
Why regular people should care
Tools like CRISPR turned odd bacterial immune systems into medicines and research staples. ART is not a ready-made gene editor. Anthropic is clear that its function is still unknown and that human scientists are doing the wet-lab validation. What is new is the workflow: AI agents combing nature’s junk drawer for patterns people might miss, then handing the best hunches to a real lab.
If even a fraction of those hunches pan out over the next few years, the path from “odd DNA repeat” to “useful biotech tool” could get shorter — with humans still holding the pipettes.
What’s next
Watch for lab results on what ART actually does, and whether other AI-led genome-mining campaigns turn up systems as programmable as CRISPR. For now, treat this as a promising lead, not a miracle cure in a bottle.
Sources: Anthropic, technical report PDF.