What was found
Anthropic said on 23 September that Claude identified a previously uncharacterised enzyme system in the DNA of bacteriophages, the viruses that infect bacteria. The company calls it array-associated reverse transcriptases, or ART.
An ART system has three parts: a reverse transcriptase enzyme, a partner gene, and an array of evenly spaced DNA repeats that resembles a CRISPR array. That resemblance is what makes it interesting. CRISPR’s repeat arrays are what make the system programmable, and a repeat array sitting next to an unusual reverse transcriptase is a structure nobody had catalogued.
The search worked at a scale that is the point of the exercise. Claude agents went through a DNA sequence database, pulled together more than 200,000 reverse transcriptases, flagged about 3,500 candidate systems and narrowed those to twenty worth looking at closely. One agent spotted the repeating pattern next to the unusual gene. Anthropic puts the total at roughly 950 agents, about 21 hours, and 210 million tokens.
What Anthropic is not claiming
The company is unusually careful about the boundaries, and the boundaries are where the story is.
Claude did not run an experiment. Every piece of wet-lab work — expressing the protein in standard laboratory strains, characterising it biochemically — was done by human scientists at Anthropic’s Bay Area lab, which opened in the spring and operates only at biosafety levels 1 and 2, with no human pathogens.
The function is not known. Early experiments showed the ART array is expressed as distinct short RNA molecules, which is what you would expect if the system were programmable in the way CRISPR is. Anthropic’s own wording is that “we don’t yet know its function.”
And it is not unprecedented. Chief executive Dario Amodei said the system was found “mostly, though not entirely, by Claude”, and acknowledged that Stanford researchers had previously described a system that is in some ways similar.
Why this counts as a result
Feng Zhang of MIT, one of the people who turned CRISPR into a tool, called it “an exciting example of how AI agents can contribute to biological discovery.” That is an endorsement of the method rather than of the enzyme, which is the right thing to endorse at this stage.
What the method did was a comparative genomics sweep at a scale a research group would not attempt by hand, followed by a judgement call about which twenty of 3,500 candidates were worth bench time. Neither half is beyond human researchers. The combination, done in a day, is new.
Anthropic has published a preprint and is inviting outside groups to propose functional studies. That invitation is the honest part of the announcement, and it is also the test. A structure that resembles CRISPR and produces short RNAs may turn out to be a programmable system, a defence mechanism with no application, or something in between. Which one it is will be settled in other people’s laboratories over the next year, not in a company blog post.
