Anthropic gave hundreds of Claude AI agents a massive biological search task. They surfaced an enzyme system scientists had not previously characterized.
The system, called array-associated reverse transcriptases, or ARTs, emerged from an AI-assisted search through more than 200,000 reverse transcriptases. Its repeating DNA structure shares some similarities with CRISPR systems, although researchers still do not know what ARTs actually do.
The findings have not undergone peer review, but the experiment offers an early look at a potentially important role for AI agents in science: searching enormous datasets, generating leads, and narrowing the candidates worth testing in the lab.
Claude searched more than 200,000 enzymes
The bigger business takeaway is not that AI can independently conduct scientific research. It is that agentic systems may dramatically expand the number of possibilities researchers can investigate before committing people, laboratory capacity, and money to physical experiments.
That could make AI particularly useful during the exploratory stage of R&D, where teams must search enormous datasets, identify unusual patterns, and decide which leads deserve further investigation.
ART also illustrates the limitation. Claude helped surface the candidate, but human researchers still had to determine whether the pattern was scientifically meaningful and test it experimentally.
One agent spotted an unusual repeating DNA pattern beside an RT gene. Claude measured the repeats, analyzed their spacing, compared the structure with known biological systems, and searched existing scientific literature for previous reports.
That finding prompted Anthropic’s researchers to investigate the candidate experimentally.
The system primarily appears in bacteriophages, viruses that infect bacteria, and includes a reverse transcriptase, a neighboring partner gene, and a long array of evenly spaced DNA repeats.
Laboratory experiments found that the repeat array is expressed as multiple short RNA molecules.
Why researchers are comparing ART with CRISPR
The repeating DNA structure is notable because CRISPR systems also contain arrays of repeated sequences.
CRISPR eventually became the foundation for programmable gene-editing tools capable of targeting specific genetic material. Researchers have also discovered other biological systems that combine reverse transcriptases with components involved in copying or manipulating DNA.
But the structural similarity does not mean ART can perform the same functions.
Anthropic said researchers still do not know ART’s biological role or whether it could eventually have biotechnology applications.
Feng Zhang, an MIT professor and Broad Institute researcher who helped pioneer CRISPR genome editing, said the connection between RNA-repeat arrays and reverse transcriptases makes the system worthy of further investigation.
That uncertainty is important. Claude found an unusual biological pattern and helped generate a hypothesis, but laboratory experiments are still needed to determine what the system actually does.
AI agents move deeper into scientific research
The experiment is notable as much for the research process as for the enzyme system itself.
Rather than asking Claude to summarize published studies or analyze a predefined dataset, Anthropic gave the system a broader research objective. Hundreds of AI agents searched databases, investigated RT families, eliminated weaker candidates, and produced reports on promising findings.
Human researchers remained responsible for evaluating those findings and conducting laboratory experiments.
Anthropic is developing this approach through its life sciences research group and laboratory, where researchers are testing whether general-purpose AI agents can participate in more stages of scientific discovery.
The company said its workflow combines Claude Science, Claude Code, and internal systems capable of coordinating multiple Claude sessions.
That model could eventually extend beyond biology. Research teams in drug discovery, chemistry, materials science, and engineering routinely search massive datasets for patterns that could point to new compounds, materials, or mechanisms.
AI agents could help researchers explore more possibilities before committing time and money to physical experiments.
What this means for organizations using AI
For businesses, the larger takeaway is not that AI can independently conduct scientific research.
Instead, the experiment shows how agentic AI could become useful in the exploratory stage of research, where teams must sift through huge amounts of information before deciding which ideas deserve further investigation.
That distinction matters for organizations evaluating AI for research and development. Systems such as Claude may be able to expand the number of hypotheses researchers can investigate, but AI-generated findings still require expert review and experimental validation.
ART demonstrates both sides of that equation. Claude identified a potentially novel biological system from a dataset that would be difficult for researchers to examine manually at comparable speed and scale, while scientists were still needed to determine whether the finding was meaningful.
For enterprises considering AI-assisted research, the near-term opportunity is therefore less about replacing researchers and more about increasing the number of promising leads they can pursue.
Whether ART becomes a useful biotechnology tool remains unknown. But its discovery offers a clearer picture of where AI agents could fit into scientific research: searching more possibilities, surfacing unexpected connections, and leaving experimental validation to humans.
Want to learn more AI tips, tricks, and prompting techniques? TechRepublic readers get free 7-day access to The Neuron Academy, our practical learning platform designed to help professionals use AI more confidently at work. Browse all lessons →
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →