Anthropic's Claude identified a previously unknown enzyme system containing CRISPR-like repeats by autonomously searching a massive database of DNA sequences, marking the first concrete demonstration of a frontier model discovering a biological system that had escaped human notice. The discovery, called array-associated reverse transcriptases (ART), emerged after Claude agents spent 21 hours combing through genomic data using roughly 950 parallel instances and 210 million tokens to narrow 200,000 reverse transcriptases down to candidates worth investigating.
The system Claude found consists of three parts: a reverse transcriptase enzyme (a protein that copies RNA into DNA), a partner gene, and an array of evenly spaced DNA repeat sequences. According to Anthropic's account, the underlying reverse transcriptase itself had been identified in previous studies in a jumbo phage, but Claude appears to be the first to recognize the defining feature—the associated array of non-coding DNA sequences and an additional accessory protein of unknown function. The repeat pattern resembles a CRISPR array in structure, which is why Anthropic named the system ART. Early experiments show the ART array is expressed as a set of distinct short RNAs, suggesting a programmable function similar to CRISPR systems, though the actual mechanism remains unknown.
Anthropic's workflow demonstrates how the model was deployed for this task. Claude agents read relevant literature, reproduced established results from public data to validate their methods, then searched for family members or genomic neighbors that fit no described system. The model wrote human-readable reports for each candidate proposing a function and describing supporting evidence. Most candidates were eliminated in follow-up analysis where Claude critically evaluated the evidence. When a candidate survived human review, Anthropic's scientists tested it in the laboratory, expressing the protein in standard strains and characterizing it biochemically and structurally, with Claude helping interpret the data.
The scale of the search reveals both the capability and the computational cost. Claude agents gathered over 200,000 reverse transcriptases, picked out 3,500 new candidate systems, and narrowed those to 20 most-compelling candidates for detailed analysis. Anthropic notes that for an expert human scientist, this type of analysis can take weeks to months of work. The discovery itself came when one Claude agent, while examining an unusual RT family in detail, detected a tandem repeat array in the raw DNA sequence and flagged it as potentially novel. The agent then counted repeats, measured spacing, compared the layout with known RT systems, and searched the literature for any previous report—all without explicit instruction to do so.
Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, reviewed the pre-print and said the identification of RNA-repeat arrays associated with reverse transcriptases is "genuinely intriguing and merits further investigation." Anthropic has released a technical pre-print with further detail and is sharing the findings early to demonstrate Claude's capabilities and give the broader scientific community insight into the work.
The hard part is what comes next. Anthropic's team is still working to understand the primary function of ARTs. The discovery shows Claude can autonomously detect anomalies and generate hypotheses at scale, but the model's role remains bounded: human scientists performed all lab work, interpreted results, and made the final judgment on which candidates were worth testing. The system works because Anthropic built a dedicated lab and a single team working on everything from training Claude in biology to running experiments, meaning the feedback loop between model output and experimental validation is tight and immediate. For teams evaluating frontier models on domain-specific tasks, the takeaway is that autonomous discovery requires not just a capable model but a closed loop where human expertise validates and directs the model's output at every stage.