950 Claude agents narrowed 200,000 enzyme genes to 20 leads in 21 hours

Twenty leads out of more than 200,000 candidate genes is a filter that keeps one in 10,000 ^[15]. Anthropic did not report an error rate for it: how many of the 20 became dead ends at the bench, or how many comparable systems sitting in the same database the agents passed over. The 21 hours measures throughput ^[4], and what a research lead would need before moving first-pass screening off a human's desk is precision and recall.

The compute side is easy to total. 950 agents for 21 hours is 19,950 agent-hours ^[16], which works out at about 1,000 agent-hours per lead ^[17]. Anthropic's stated alternative for the same job is weeks or even months of an expert scientist's time ^[21], and the company supplies that estimate itself.

What the run shows is volume at the screening stage. Anthropic said its researchers' involvement was limited to the initial instructions and the lab experiments, with the agents deciding autonomously which leads to pursue ^[6].

The find sits in well-trodden territory. Repeat arrays beside enzyme genes are the layout of CRISPR ^[2]. That layout is easy for a search to recognise and hard to call new until someone knows what it does. Anthropic said the function of the system, which it dubbed ART, remains unknown ^[7], and that initial experiments suggest it could be programmable ^[20].

Feng Zhang, one of the pioneers of CRISPR and a professor at MIT and the Broad Institute, called the identification of the repeated DNA fragments "genuinely intriguing" and said it "merits further investigation," in remarks released by Anthropic ^[12]. Amodei said on X that a team at Stanford University had independently described a comparable but distinct system ^[13]. A distinct system supports the idea that this family of arrangements is real without replicating ART.

Amodei, who recently called for slowing AI development because of its risks ^[14], cited the discovery in remarks to the UN Security Council delivered by video link ^[11]. The finding is "preliminary," he said, but "may constitute a new gene editing mechanism that could have applications in gene therapy" ^[18]. He also predicted "profound progress in the near future on diseases that have plagued humanity for millennia" ^[19].

Anthropic's lab team was formed in the spring, works in the San Francisco Bay Area, and does not handle pathogens capable of infecting humans ^[9]. All experiments are carried out by human scientists, the company said, and Amodei did not rule out Claude eventually operating lab equipment itself, with appropriate safeguards ^[10].