Anthropic has formed a new life sciences research group and wet lab where its AI model Claude autonomously identified a previously uncharacterized enzyme system with structural patterns reminiscent of CRISPR. The discovery, announced on September 23, 2026, came after Claude agents spent 21 hours searching a massive DNA sequence database and flagged a repeating array of non-coding DNA next to a reverse transcriptase gene in a bacteriophage.
The research group, part of Anthropic's broader life sciences organization, combines computational biology with hands-on laboratory experiments. The team gave Claude a high-level prompt to search for interesting reverse transcriptase (RT) variants and then stepped back while roughly 950 AI agents using 210 million tokens combed through over 200,000 RT sequences. From those, the agents narrowed candidates to 3,500 new systems and produced human-readable reports on the 20 most compelling ones.
"This is an exciting example of how AI agents can contribute to biological discovery," said Feng Zhang, a CRISPR pioneer and professor at MIT and the Broad Institute, after reviewing the pre-print. "The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research."
How Claude spotted the pattern
During its analysis, one Claude agent noticed an unusual RT family and examined the surrounding raw DNA sequence. The agent reported seeing a tandem repeat array by eye, comparing it to CRISPR-like structures. It then counted the repeats, measured their spacing, cross-referenced known RT systems, and searched the literature for any prior description of the pattern. After confirming no previous reports existed, it filed a report for human review.
The discovered system, which the team calls array-associated reverse transcriptases (ART), consists of three components: the RT enzyme itself, a partner gene adjacent to it, and a long array of evenly spaced DNA repeat sequences. This repeat layout resembles a CRISPR array, which stores distinct RNA sequences that make CRISPR-Cas systems programmable. Initial lab experiments show the ART array is expressed as a set of distinct short RNAs, suggesting a similar programmable mechanism may be at work.
From computational search to wet lab validation
The team's workflow follows a consistent pattern. Claude reads relevant literature and reproduces established results from public data to verify its methods. It then searches for protein family members or genomic neighbors that fit no described system and writes short reports proposing functions with supporting evidence. In follow-up analyses, Claude critically evaluates the evidence, and most candidates are eliminated at this stage. When a candidate survives review, human scientists test it in the lab, expressing the protein in standard strains and characterizing it biochemically and structurally, with Claude helping interpret the resulting data.
Because Claude generates hypotheses at scale-hundreds to thousands of candidate reports from a single campaign-the team studies the proposals themselves to understand what distinguishes the ones worth testing. Those insights feed back into the instructions given to Claude, teaching it to mirror the scientists' own judgment. For a human expert, the type of analysis Claude performed can take weeks to months.
What ART means and what comes next
The underlying RT enzyme, found in a jumbo phage, had been identified in previous studies. Claude appears to be the first to notice the defining features: the associated array of non-coding DNA sequences and an additional accessory protein of unknown function. The system's characteristics have only been found together in a handful of other systems, all of which are programmable and perform operations like cutting, copying, and pasting DNA. Several such systems beyond CRISPR are now in development as biotechnological tools.
The team does not yet know ART's primary function, and further experiments are underway. The lab, located in the Bay Area, operates at biosafety levels 1 and 2 and does not handle human pathogens. All lab work is performed by human scientists using standard molecular biology techniques. The research group has released a pre-print with full technical details and is seeking collaborators who have proposals for research questions in genomics or other fields.
Why this matters for science and research professionals
This result demonstrates a shift in how biological discovery workflows can operate. Rather than a scientist manually noticing an anomaly in sequence data, Claude agents performed the entire search, candidate triage, and hypothesis generation autonomously, with humans stepping in only for the initial prompt and final lab validation. For researchers in genomics, drug discovery, or molecular biology, the implication is that AI-driven hypothesis generation can compress months of expert analysis into hours of computation. Professionals looking to build these skills can explore structured learning paths such as AI Scientific Research Courses that cover computational approaches to biological data. The approach also raises a practical question for research teams: what changes when hypotheses become cheap and abundant, and human judgment becomes the primary bottleneck.
Your membership also unlocks:
