Anthropic said on 23 September that AI agents working in its new life sciences research group identified a previously uncharacterised enzyme system in a jumbo bacteriophage: a reverse transcriptase sitting beside an array of evenly spaced non-coding DNA repeats and an accessory protein of unknown function. The repeat array is the arresting detail, because that is the pattern which first drew attention to CRISPR decades ago.
The figures Anthropic gives for the search: roughly 950 agents, 21 hours and 210 million tokens spent combing a sequence database, gathering more than 200,000 reverse transcriptases, narrowing those to 3,500 candidate systems and then to 20 written up as reports for human review.
The caveats are the story as much as the finding. The underlying enzyme had been identified in earlier studies; the claim is that Claude was first to notice the system’s defining features together. What the system actually does is unknown. All laboratory work was carried out by human scientists at biosafety levels 1 and 2. Feng Zhang of MIT and the Broad Institute, who reviewed the preprint, called the result intriguing and worth further, which is an endorsement of the method, not yet a verdict on the biology.