Claude Discovers Novel CRISPR-like Enzyme System in Phage DNA
Anthropic announced that its Claude AI discovered a novel CRISPR-like enzyme system, named ART, in phage DNA. The discovery was led by 950 Claude agents operating over 21 hours, identifying a repeating array structure. CRISPR pioneer Feng Zhang called it a strong example of AI-driven biological discovery.
Key insights
The Discovery of the ART System
Anthropic announced that Claude discovered a new enzyme system in phage DNA, which they named ART (Array-associated Reverse Transcriptase). The system consists of a reverse transcriptase, a partner gene of unknown function, and a long array of evenly spaced repeating DNA sequences, structurally resembling CRISPR. Experiments confirmed the array is transcribed into distinct short RNAs, suggesting a potentially programmable mechanism.
AI-Driven Discovery Process
The discovery was almost entirely led by Claude. Researchers provided a single initial prompt to search for interesting new reverse transcriptases. Approximately 950 Claude agents worked for 21 hours, consuming 210 million tokens. They screened over 200,000 reverse transcriptases, narrowed them down to 3,500 candidates, and then to 20 top candidates. One agent independently recognized the repeating array in the raw sequence data.
Significance and Expert Endorsement
The discovery is significant because systems with these combined features are rare and have historically led to powerful biotech tools like CRISPR. CRISPR pioneer Feng Zhang from MIT reviewed the preprint and stated it is a strong example of AI agents participating in biological discovery, emphasizing that such arrays deserve further study.
Anthropic's Life Sciences Initiative
The work was conducted by Anthropic's newly formed life sciences team, which operates its own BSL-1 and BSL-2 wet lab in the Bay Area. The team uses publicly available tools like Claude Science and Claude Code, meaning other labs can theoretically replicate the workflow. Anthropic is actively soliciting research collaboration proposals from external scientists.
Comparison to Prior and Concurrent Work
Dario Amodei noted that this discovery builds on decades of prior work, including CRISPR, bridge recombinase, and VIPR. He also acknowledged that a Stanford team independently described a novel RT system with a non-coding array that is similar in some ways but distinct and evolved independently from the one Claude found.
Exponential Trend in AI for Biology
Dario Amodei drew a parallel to AI's trajectory in mathematics, progressing from high-school level in 2023 to solving major open problems in 2026. He believes AI for biology is on a similar exponential trend. While biology requires experimentation, he argues that AI can collaborate with humans to perform experiments and validate results, eventually potentially controlling lab equipment autonomously.
Limitations and Next Steps
Anthropic acknowledged that the precise biological function of the ART system is currently unknown, and its biotechnological utility is not yet clear. All experiments were physically performed by human scientists. The long-term goal is to accelerate the entire biomedical pipeline from fundamental discovery to drug delivery, with the aim of curing most diseases within 5-10 years.