Claude discovers CRISPR-like enzyme system ART in phages

Anthropic announced that Claude identified a previously unknown enzyme system in phage DNA, named ART (Array-associated Reverse Transcriptase), which features CRISPR-like repeat arrays. The discovery was made by 950 Claude agents running for 21 hours, consuming 210 million tokens, and was validated by human scientists.
Key insights
ART: A New CRISPR-Like Enzyme System
Anthropic's Claude discovered ART (Array-associated Reverse Transcriptase) in phage DNA. The system comprises a reverse transcriptase, an unknown partner gene, and a long array of evenly spaced repeated DNA sequences, structurally resembling CRISPR. The array is transcribed into distinct short RNAs, suggesting a programmable mechanism similar to CRISPR.
Discovery Process: 950 AI Agents in 21 Hours
Claude conducted the discovery autonomously. Researchers gave an initial prompt to search for interesting new reverse transcriptases in DNA databases. 950 Claude agents ran for 21 hours, consuming 210 million tokens. They filtered over 200,000 reverse transcriptases to 3,500 candidates, then selected the top 20 for human-readable reports. One agent recognized the repeat array in raw sequence data, counted repeats, measured spacing, compared with known systems, and confirmed it was novel.
Significance: Potential Gene Editing Tool
Only a handful of systems with these features have been found, and all are capable of DNA cutting, copying, or pasting. CRISPR has already led to gene-editing drugs, and other similar systems are being developed into tools. Anthropic notes that ART's exact function is still unknown, but its structural similarity to CRISPR suggests it may be programmable. CRISPR pioneer Feng Zhang called it a good example of AI agents contributing to biological discovery.
Human Validation and Wet Lab Work
Anthropic's life sciences research group, formed in spring 2026, conducted all experiments in their own wet lab in the Bay Area, limited to BSL-1 and BSL-2 low-risk research. Human scientists performed the lab work to validate Claude's findings. The team uses publicly available Claude Science and Claude Code tools, meaning other labs could replicate the AI-driven hypothesis generation and experimental validation pipeline.
Implications for AI-Driven Scientific Discovery
This discovery demonstrates that AI agents can independently identify novel biological systems from raw genomic data, a task that typically takes human experts weeks to months. The process involved pattern recognition, literature comparison, and report generation. Anthropic is now seeking collaboration with external scientists on research topics.