What was the extent of human involvement in Claude's discovery of the ART system?

Human scientists at Anthropic initiated the project by giving Claude a high-level prompt to search a DNA database for interesting reverse transcriptases. However, the subsequent tasks—sifting through the data, identifying thousands of candidates, analyzing the unique repeat pattern, and generating the initial report—were performed autonomously by Claude agents. Humans then reviewed the AI's findings and conducted the physical laboratory experiments to verify the existence and characteristics of the ART system.

Sep 23, 2026ScienceClaude discovers a novel enzyme system with CRISPR-like repeats

The News

Anthropic has established a new life sciences research group and laboratory, announcing that its AI model, Claude, autonomously discovered a novel enzyme system with features similar to CRISPR. The discovery of what the company calls array-associated reverse transcriptases (ART) was achieved with only a high-level prompt from human scientists. This finding serves as an early proof-of-concept for a new research methodology where AI agents actively participate in generating and analyzing scientific hypotheses from massive biological datasets.

The Details

The project began when researchers tasked Claude with searching a vast DNA sequence database for interesting new examples of reverse transcriptases (RTs), enzymes that copy RNA into DNA. A coordinated effort involving roughly 950 Claude agents running in parallel for 21 hours combed through the data, ultimately analyzing 3,500 new candidate systems. One agent identified a remarkable pattern: a repeating array of DNA sequences located next to an unusual RT gene found in bacteriophages, which are viruses that infect bacteria. This structure, consisting of the RT, a partner protein, and the DNA array, is reminiscent of other programmable systems used in biotechnology.

The technical specifics of the AI-driven search highlight the scale of the operation:

AI Model: Claude, operating as a team of coordinated agents.
Task: Identify novel reverse transcriptase (RT) systems from public genomic data.
Scale: Approximately 950 agents ran for 21 hours, using 210 million tokens.
Outcome: Discovered a previously uncharacterized system named array-associated reverse transcriptases (ART).
Human Role: Provided the initial high-level prompt and performed subsequent wet lab verification.

The Impact

This work demonstrates a significant step in using AI not just for data analysis, but for active hypothesis generation in fundamental biology. By building its own Bay Area lab, Anthropic is creating a tight feedback loop where AI-generated ideas can be rapidly tested, potentially compressing research timelines from months or years into days. While the function of ART is still under investigation, the methodology itself signals a potential shift in how biological discovery is conducted. This model of human-AI collaboration could allow scientists to more systematically explore the uncharacterized majority of proteins and genes in nature.

Anthropic is not just reporting a biological finding; it's demonstrating a scalable, AI-driven discovery engine. By integrating large-scale computation with its own wet lab for verification, the company is building a closed-loop system where AI generates novel hypotheses and humans validate them, moving AI from a passive analysis tool to an active participant in the scientific process.

>> Verify Original Transmission at Anthropic