Claude agents identify an uncharacterized CRISPR-like enzyme system
On 23 September 2026 Anthropic announced a life sciences research group and its own laboratory, and released a preprint describing the group's first result. Claude agents searching a DNA sequence database ran autonomously for roughly 21 hours across about 950 agent sessions, consuming about 210 million tokens; they assembled roughly 200,000 reverse transcriptases, narrowed these to about 3,500 candidate systems, and produced about 20 reports for human review. One agent noticed a repeating array of DNA sequences sitting next to the gene for an unusual reverse transcriptase in bacteriophage DNA, together with a partner protein of unknown function. Anthropic named the combination array-associated reverse transcriptases (ART) and said it resembles the repeat structure underlying CRISPR. The company stated it has not determined what ART does, or whether it can be used for gene editing. The reverse transcriptase itself had been described in earlier work; what Claude appears to have found first is the wider system. Feng Zhang of MIT and the Broad Institute reviewed the preprint and said the RNA-repeat arrays associated with reverse transcriptases merit further investigation.
Why It Mattered
This is a different mode of AI contribution to biology than structure prediction or literature synthesis. AlphaFold-class systems answer a question a scientist poses; here the agents were given a high-level instruction to look for unusual reverse transcriptases in a large sequence database and chose for themselves what counted as anomalous, following the anomaly through comparison and characterisation before handing reports to humans. Historically, restriction enzymes, Taq polymerase and CRISPR itself all began as an oddity someone noticed in sequence or culture data, so the step that was automated here is the one that has repeatedly preceded major biotechnology. Two cautions belong in any 2030 account. First, the function of ART is unknown; the CRISPR comparison rests on architecture, not demonstrated programmability, and Anthropic said so explicitly rather than claiming a gene-editing tool. If characterisation fails, the entry becomes a methodological data point rather than a discovery. Second, the result is a preprint from the company that built the model, not peer-reviewed independent work, though an outside CRISPR researcher's qualified endorsement raises its standing. The institutional fact may outlast the biological one: a frontier model developer now runs a wet lab and publishes biology preprints, placing hypothesis generation and experimental validation inside the same organisation that trains the model. That vertical arrangement, if copied, changes where biological discovery happens and who owns it.
Who Built It
Anthropic
Applications
- Molecular Biology
- Gene Editing
- Scientific Discovery