Claude Finds CRISPR-Like Enzyme System in 21-Hour DNA Search

On September 23, Anthropic published a post saying its Claude model had autonomously identified a previously uncharacterized enzyme system in the DNA of bacteriophages, the viruses that infect bacteria. The company calls it array-associated reverse transcriptases, or ART, and says its layout is reminiscent of CRISPR.

ART has three parts, according to the post: a reverse transcriptase, the enzyme that copies RNA into DNA; a partner gene of unknown function beside it; and a long array of evenly spaced DNA repeats. The reverse transcriptase itself, found in a jumbo phage, had already been described in earlier studies; what is new, Anthropic says, is that Claude appears to be the first to notice the array and the accessory protein together.

The search ran for 21 hours. Anthropic's technical report states that roughly 950 agents used about 210 million tokens across a database of 1.9 billion metagenomic protein clusters. The agents gathered more than 200,000 reverse transcriptases, isolated 3,500 candidate systems and narrowed those to the 20 most compelling, each written up for human review. The report quotes the agent that hit the pattern: "[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array ... that's a CRISPR-like ... repeat array?!"

Anthropic also published what did not work. It ran the same campaign ten more times, and the report states the array was missed in every rerun, which it attributes to the breadth of the search space and the non-deterministic behaviour of its agent harness. In fixed-input tests, the most capable models described the array in at least 90% of attempts with the sequences directly in context, but in as low as 32% when working through files and tools. Counted with the original campaign, that is one array found in eleven full runs, against at least nine attempts in ten when the DNA was placed straight in front of the model. Anthropic's own diagnosis points the same way: 39% of the file-based attempts never read 200 consecutive nucleotides.

What ART does remains unknown. Anthropic says first experiments show the array is transcribed into a set of distinct short RNAs, and that work on its primary function continues. Feng Zhang of MIT and the Broad Institute, a CRISPR pioneer, said on Anthropic's page that the finding is "an exciting example of how AI agents can contribute to biological discovery". Stanford bioengineer Stanley Qi told Al Jazeera it was "incredibly exciting". Kevin Blake, a microbiologist at Washington University School of Medicine, was blunter to the same outlet: "There's nothing to indicate this is a rival to CRISPR-the-technology, or could be developed into any kind of therapeutic or practical application."

Chief executive Dario Amodei wrote on X, per Al Jazeera, that the company suspects the machine could "represent a new gene editing mechanism" - a suspicion, not a result. Next come the lab experiments Anthropic says are under way, and whether outside groups reproduce the preprint's analysis.

Disclosure: NewUJ's editorial process uses Anthropic's Claude models.