Claude Finds CRISPR-Like Enzyme System in DNA Search

Anthropic says Claude agents have helped identify a previously uncharacterized enzyme system in bacteriophage DNA, with a structure that resembles some of the programmable biology associated with CRISPR. The company is presenting the result as an early example of AI-assisted scientific discovery, not as a finished gene-editing breakthrough, according to Anthropic’s announcement.

What Claude found

The system, which Anthropic calls array-associated reverse transcriptases, or ART, combines a reverse transcriptase with a neighboring partner gene and a long array of repeated DNA sequences. Reverse transcriptases copy RNA into DNA. The repeat array is interesting because it is reminiscent of the arrays that help make CRISPR systems programmable, although Anthropic says it does not yet know what ART does.

The discovery came from a large genome-mining campaign. Anthropic says roughly 950 Claude agents searched for 21 hours, processing about 210 million tokens. The agents gathered more than 200,000 reverse transcriptases, narrowed them to 3,500 candidate systems, and produced a smaller set of reports for human review. One agent then noticed a repeat pattern beside an unusual reverse transcriptase and flagged it for laboratory investigation.

The important caveat

Human scientists performed the experiments, and the biological function of ART remains unknown. The Verge notes that it is still unclear whether the system will have practical applications, despite the comparison to CRISPR and the excitement around AI-driven biology (The Verge). Anthropic has also released the result as early work while further experiments continue.

Builder impact

For AI builders, the useful signal is the workflow rather than the headline. The system combined broad search, parallel agent sessions, literature checks, candidate ranking, human review, and physical validation. That is a pattern for high-value agentic research: let models expand the search space, but keep evidence thresholds, review gates, and real-world tests outside the model’s authority.

The next test is reproducibility. If researchers can repeatedly recover meaningful candidates and explain why they survive review, multi-agent systems may become a practical layer for scientific exploration. Until then, ART is a promising lead, not a proven biotechnology platform.