Anthropic announced on Wednesday that its artificial intelligence model Claude autonomously identified a previously undescribed enzyme system with features similar to CRISPR. The company published the findings in a pre-print and established a new life sciences research group and laboratory to test whether AI can accelerate biological discovery. The newly named system, called ART for array-associated reverse transcriptases, appears mainly in bacteriophages and combines three distinctive components.
A New Research Effort in AI-Driven Biology
Anthropic's new laboratory in the San Francisco Bay Area focuses on fundamental biology research by using Claude to explore DNA datasets and identify uncharacterized protein families. The facility operates only at lower biosafety levels and does not handle pathogens that can infect humans. All laboratory work is performed by human scientists, with Claude assisting in data interpretation and experimental analysis.
How Claude Conducted the Genomic Search
Scientists gave Claude a single instruction to search a massive DNA sequence database for interesting examples of reverse transcriptases, enzymes that copy RNA into DNA. Claude agents gathered more than 200,000 sequences, selected 3,500 new candidate systems, and narrowed the list to the 20 most promising cases for written reports. The search lasted 21 hours, used about 950 agents running in parallel, and consumed 210 million tokens of model processing.
The ART System and Its CRISPR-Like Pattern
During the search, one agent noticed an unusual pattern of repeated DNA sequences next to the gene for an odd-looking reverse transcriptase. The agent counted the repeats, measured their spacing, compared the layout with known systems, and searched scientific literature before filing a report for human review. The system includes a reverse transcriptase, a neighboring gene of unknown function, and an array of evenly spaced DNA repeats that had not been linked together.
Known Enzyme, New Context
Although the underlying reverse transcriptase had been identified in previous studies, the associated repeat array and accessory protein had not been connected to it. The newly recognized ART system is found mainly in bacteriophages, and its repeat layout resembles a CRISPR array. Initial experiments show that the ART array is expressed as a set of distinct short RNAs, suggesting that an analogous programmable mechanism may be at play.
Scientific Reception and Reproducibility Challenges
Feng Zhang, a CRISPR pioneer and professor at MIT and the Broad Institute, reviewed the pre-print and called the identification of RNA-repeat arrays associated with reverse transcriptases genuinely intriguing. Anthropic acknowledged that the result is not easily reproducible, because repeating the same search ten times did not lead agents back to the relevant DNA segment. In controlled tests where researchers provided the relevant DNA directly, the company's models identified the repeat pattern in at least 90 percent of attempts.
Strategic Context and Executive Caution
The announcement is part of a broader push by Anthropic into life sciences, following its reported US$400 million acquisition of biotechnology startup Coefficient Bio in April. The company also launched Claude Science in June, a platform that integrates more than 60 scientific databases. Chief Executive Officer Dario Amodei commented cautiously, saying the precise function, biotechnological utility, and significance remain unclear but that the work is something he would be proud of.
Anthropic's early findings suggest that general AI models can autonomously detect biological anomalies and initiate meaningful discovery workflows, even if the long-term value of ART is still unknown. The company has released a technical pre-print and says it wants to collaborate with other scientists on similar research questions. Further experiments are underway to determine how ART works and whether it can be developed into a programmable biological tool.