Not reviewed. This note reports what the post itself says. Unlike the other notes in this collection, its claims have not been checked against the underlying papers, reports or data. The work concerns biological design that could be misused, and in-depth analysis of it triggers the safety safeguards of the AI system used to prepare these notes, so it was left at the level of the post's own summary.
What the post reports
The post introduces a life sciences research group and laboratory formed at Anthropic in the spring of 2026, whose subject is fundamental biology: reading large collections of DNA sequence to find protein families nobody has characterized, generating hypotheses at scale, and testing the survivors in the lab. Its framing is historical, in that restriction enzymes, Taq polymerase and CRISPR all began with someone noticing something odd in natural diversity, and the question is whether a general AI model can systematize that kind of noticing.
The early result is a previously uncharacterized enzyme system the team calls ART, array-associated reverse transcriptases, found mainly in bacteriophages. The post says scientists supplied only the initial prompt and the lab work; Claude agents searched the database, examined the families they found, and used their own judgement about what was interesting. It reports roughly 950 agents spending 21 hours and 210 million tokens, gathering over 200,000 reverse transcriptases, picking out 3,500 candidate systems and narrowing those to the 20 most compelling, each written up as a human-readable report, work the post says would take an expert weeks to months. One agent noticed a repeating pattern beside an unusual gene and, in a quoted exclamation, recognized it as a repeat array of a kind reminiscent of CRISPR. The post says the underlying enzyme had been identified in earlier studies, but that Claude appears to be the first to notice the system's defining features, and that its combination of characteristics has been found together only in a handful of other systems, all of them programmable. First experiments, it says, show the array is expressed as a set of distinct short RNAs, suggesting something analogous may be at work.
The post also describes how the group works: Claude reads the literature, reproduces established results from public data as a check on its methods, proposes candidates with supporting evidence, then critically evaluates them, and a survey may end with one candidate worth testing, or none. Because the hypotheses arrive in such volume, the post says the group has begun studying what distinguishes the proposals it judges worth testing, and feeding that back into Claude's instructions.
Who did the work and who checked it
The authors are Anthropic's new life sciences research group, described as scientists whose earlier careers were spent on CRISPR evolution and regulation, new enzymes for cell and gene therapies, and tools for finding anomalies in DNA. The detail is in a preprint on bioRxiv, so peer review is not claimed. Feng Zhang, a CRISPR genome-editing pioneer at MIT and the Broad Institute, reviewed the preprint and is quoted calling the finding "genuinely intriguing" and saying it "merits further investigation."
Caveats the post itself gives
The function of the system is not known: the post says work to understand it is ongoing, that the RNA result only suggests an analogy, and that it is sharing early deliberately, to demonstrate Claude's capabilities and show the community what the group is working on. On safety, the post states the lab works only at the lower biosafety levels (BSL-1 and BSL-2), does not handle pathogens that can infect humans, and that all lab work is performed by human scientists; it adds that automating the bench suits its ad hoc workflows less well.
How it connects
- Teaching AI to build scientific models is listed beside it as related work from the same program.
- Introducing our Science Blog describes the Science blog through which this lab publishes.
References
- "Claude discovers a novel enzyme system with CRISPR-like repeats." Anthropic Science blog, 23 September 2026. https://www.anthropic.com/research/claude-discovers-novel-enzyme-system record