Anthropic has disclosed one of its most unusual demonstrations of Claude yet: an AI-assisted biology project in which the company says its agents identified a previously uncharacterized enzyme system with features reminiscent of CRISPR.
The company says the discovery came from a new life sciences research group and laboratory established in 2026. Researchers gave Claude a high-level objective to search large DNA databases for interesting examples of reverse transcriptases. The agents then examined candidate sequences, compared them with known systems and generated reports for human scientists.
Anthropic says roughly 950 Claude agents worked for 21 hours and consumed about 210 million tokens during the search. The agents examined more than 200,000 reverse transcriptases, identified about 3,500 candidate systems and narrowed the field to 20 especially compelling candidates. One of those candidates contained a pattern that prompted further investigation.
The clue was in the DNA around the enzyme
The core enzyme involved is a reverse transcriptase, a class of proteins that can copy RNA into DNA. The reverse transcriptase itself was not unknown. Anthropic says the enzyme had already been identified in a jumbo bacteriophage in earlier research.
What Claude appears to have noticed was the surrounding structure. The agent detected an array of repeated non-coding DNA sequences near the reverse transcriptase and an additional accessory protein whose function is not yet understood. Anthropic named the resulting system array-associated reverse transcriptases, or ART.
The comparison with CRISPR comes from the repeat structure. CRISPR systems contain arrays of repeated DNA sequences that can form part of programmable defence mechanisms and have become the basis for important gene-editing technologies. Anthropic is not claiming that ART is another CRISPR system. The company says its function remains unknown and that laboratory work is continuing.
That distinction is important. Finding an unusual pattern is not the same as proving what a biological system does. Anthropic’s researchers tested the candidate in the laboratory after Claude identified it, and early experiments indicated that the ART array is expressed as a set of distinct short RNAs. More experiments are needed to determine the system’s primary biological function.
The workflow is the real story
The most interesting part of the announcement may be the workflow rather than the name of the enzyme system. Biological discovery often involves searching enormous databases for unusual sequences, comparing them with existing literature and deciding which candidates deserve laboratory time. Much of that work is repetitive and difficult to scale manually.
Anthropic’s experiment treated Claude as a research workforce operating under human direction. Scientists set the initial objective and conducted the laboratory testing, while the agents handled much of the computational exploration. The company says the system produced hundreds or thousands of candidate reports, forcing researchers to develop methods for deciding which AI-generated hypotheses were worth testing.
That changes the economics of early-stage research. If an AI system can reduce a search that might take weeks or months to a smaller set of candidates within hours, laboratories can spend more of their time on experiments that require physical equipment and expert judgement.
Experts are still required at the critical point
Anthropic’s own description makes clear that the work was not a fully automated laboratory discovery. Human researchers supplied the initial direction, evaluated the output and performed experiments. The company also says it does not yet know what ART does.
Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, said the identification of RNA-repeat arrays associated with reverse transcriptases was intriguing and deserved further investigation. His response reflects the appropriate scientific status of the result: interesting enough to study, but not yet a demonstrated biotechnology tool.
The wider significance is therefore about how AI changes the search phase of biology. Researchers already use computation to mine genomes, predict structures and compare sequences. More capable agents could turn those tools into an iterative research process in which thousands of hypotheses are generated, filtered and refined before a smaller number reach the laboratory.
Whether that approach produces practical medicines, new gene-editing systems or other useful biological technologies will depend on what ART ultimately does and whether similar workflows can repeatedly produce discoveries. For now, Anthropic has demonstrated a concrete example of AI moving beyond summarising scientific literature and into the much messier process of finding something that researchers did not previously recognise.