Anthropic gave Claude a broad research task: search DNA databases for interesting enzymes. One agent spotted an unusual repeat pattern and identified a biological system that scientists had not previously characterized.
Anthropic calls the system array-associated reverse transcriptases, or ARTs. Its repeating DNA resembles part of the CRISPR system, although researchers still need to determine what role ART plays and whether it has any potential for gene editing.
The more interesting part is how the discovery happened. Claude helped decide which unusual pattern deserved attention, then human scientists took the lead in the lab.
Claude had room to decide what looked promising
Anthropic says about 950 Claude agents searched for roughly 21 hours and used 210 million tokens. They collected over 200,000 reverse transcriptases, which are enzymes that copy RNA into DNA. From there, the agents identified about 3,500 candidate systems and narrowed the list to roughly 20 for closer review.
One agent noticed repeating DNA next to an unusual reverse transcriptase gene. Researchers asked Claude to find interesting enzyme systems, but they didn't specifically ask it to search for this type of repeat pattern.
The agent reviewed the spacing of the repeats, compared the system with known biology, and searched published research before sending the finding to human scientists.
Claude’s role went beyond searching the dataset. It also helped decide which finding deserved further investigation, a capability that could become more valuable as AI systems generate more research leads than scientists have time or lab capacity to test.
The CRISPR comparison has limits
ART mainly appears in bacteriophages, which are viruses that infect bacteria. The system includes a reverse transcriptase, a nearby partner gene, and a long sequence of evenly spaced DNA repeats.
The reverse transcriptase itself was already known. As Reuters reported, earlier studies had identified the enzyme. Anthropic’s discovery claim centers on Claude recognizing that the repeat array and nearby protein appear to form a larger biological system.
The repeated DNA invites comparisons with CRISPR because CRISPR arrays produce RNA sequences that help identify biological targets. Anthropic’s early experiments found that ART’s repeat array also produces short RNAs, although researchers still need to determine their function.
CRISPR researcher Feng Zhang described the finding as “intriguing and merits further investigation” after reviewing the preprint. His comment supports further research, but it does not establish that ART will have a similar use to CRISPR.
Human scientists still control the lab work
Anthropic formed its life sciences research group in spring 2026 and built a molecular biology lab in the Bay Area. Human scientists perform the physical experiments, while Claude searches genomic data and proposes candidates for review.
The ART research remains a preprint and has not completed peer review. Anthropic is also describing the performance of its own AI system, so outside researchers will need to test whether similar results can be reproduced.
The project follows Anthropic’s earlier work on Claude for protein design and chemistry. Its in-house lab lets the company test whether AI-generated ideas produce results in real experiments.
AI-directed experiments have a track record
Researchers have built AI systems that propose experiments and use those results to decide what to test next. A 2024 protein-engineering study introduced SAMPLE, a system that proposed protein designs, sent them to automated lab equipment, and used the results to choose what to test next.
SAMPLE worked toward a specific goal: creating enzymes with better heat tolerance. Anthropic gave Claude a more open-ended task and allowed the agents to search for unusual biological systems.
The main difference is where human judgment enters the process. SAMPLE automated an experimental cycle.
Anthropic gives Claude more freedom during the search, while human scientists still decide which findings should proceed to lab testing.
AI-generated research leads could outpace lab capacity
The Neuron has also covered AI systems that connect biological analysis with experimental testing. ART pushes another part of the research process toward automation: deciding what scientists should investigate next.
If AI agents can generate credible research leads faster than labs can test them, experimental capacity could become the main constraint. Research teams may need better ways to rank machine-generated ideas and decide which ones deserve lab time.
ART alone does not provide enough evidence to show whether Claude can make reliable research decisions across different projects. Scientists still do not know what the system does, and many AI-generated leads may fail once they reach the lab.
For Anthropic, the next test is less about finding another unusual enzyme. The company now needs to demonstrate how consistently Claude can identify research leads that justify further scientific testing.
