Artificial intelligence is increasingly moving beyond chatbots, image generation and coding. Anthropic has now shared an unusual example of AI being used directly in scientific research, with its Claude model helping researchers identify a previously uncharacterized biological system hidden inside a large collection of DNA data.
The finding was announced on September 23, 2026, alongside the introduction of Anthropic’s new life sciences research group and laboratory. The discovery is still at an early stage, but it provides an interesting glimpse at how AI agents could eventually work alongside scientists to explore datasets that would take humans much longer to examine manually.
Claude was asked to explore a huge DNA dataset
The project began with a relatively broad task. Researchers asked Claude to search through large collections of DNA sequences and look for unusual examples of enzymes known as reverse transcriptases.
Instead of giving the model a specific pattern to find, Anthropic says multiple Claude agents were allowed to explore the data, compare different groups of sequences and decide which candidates appeared interesting enough for further investigation. The AI system gathered more than 200,000 reverse transcriptase sequences and reduced them to thousands of potential candidates.
It eventually selected a much smaller group for deeper analysis. According to Anthropic, the process involved around 950 AI agents working for approximately 21 hours and using about 210 million tokens.
An unusual repeating pattern caught Claude’s attention
During the search, one Claude agent noticed an unusual repeating pattern located close to one of the reverse transcriptase genes. The structure appeared interesting because the repeated DNA arrangement shared some visual similarities with the repeat patterns associated with CRISPR systems.
Claude then examined the spacing of the repeats, compared the surrounding structure with previously known systems and searched existing scientific literature to determine whether the same arrangement had already been described. The system was eventually passed to human researchers for further review and laboratory testing.
Anthropic has named the newly identified system array-associated reverse transcriptases, or ARTs.
The discovery has similarities to CRISPR, but it is not CRISPR
The comparison with CRISPR is one of the most interesting parts of the research, but it is also important not to overstate what has been found. ARTs contain an enzyme, a nearby partner gene and a long collection of regularly spaced DNA repeats.
That arrangement has some characteristics that resemble biological systems scientists already know can perform programmable operations involving genetic information. However, researchers do not yet know the primary function of ARTs.
That means the discovery should not currently be described as a new replacement for CRISPR or as a finished gene-editing technology. Instead, it is a newly identified biological system that researchers now have to understand through further experiments.
Humans still performed the laboratory work
Although Claude played a major role in exploring the data and identifying the candidate, the project was not completely automated. Human scientists reviewed the findings and carried out the physical laboratory experiments used to study the system.
Anthropic says all laboratory work at its facility is performed by human researchers. This distinction is important because it shows one possible future for AI in science.
Instead of replacing researchers, AI agents could search enormous datasets, generate possible explanations and highlight unusual patterns while scientists decide which ideas are worth testing in the real world.
AI can explore far more possibilities at once
One of the reasons AI could become valuable for scientific research is scale. Modern biological databases contain extraordinary amounts of information.
Researchers may know that valuable patterns exist somewhere inside that data, but manually examining every possible candidate can require significant amounts of time. Anthropic says the kind of genomic analysis performed during this project could take an expert scientist weeks or months.
AI agents can instead investigate many possibilities in parallel. Most of those possibilities may lead nowhere, but the ability to rapidly narrow a large dataset into a small collection of interesting candidates could give researchers more time to focus on experiments and interpretation.
Anthropic is building a dedicated AI research lab
The ART discovery is also notable because it comes from Anthropic’s own new life sciences laboratory in the San Francisco Bay Area. The company created the research group in 2026 to explore whether general-purpose AI models could help accelerate fundamental biological research.
Its researchers are using Claude to examine DNA, search for unusual protein families, generate scientific hypotheses and help interpret experimental results. The physical experiments remain in the hands of scientists, while Claude is primarily used for data exploration and reasoning.
This creates a workflow where AI and researchers repeatedly pass information between computational analysis and laboratory experiments.
Scientific AI may be about more than specialized models
Many AI systems built for science are designed around one specific task. Anthropic’s approach is somewhat different because Claude is a general-purpose AI model that is also being asked to perform scientific research tasks.
The same underlying type of model that can work with documents or software can also read scientific literature, examine biological data and generate hypotheses. If that approach continues to improve, general AI agents could become useful tools across multiple research fields rather than requiring a completely different AI system for every scientific problem.
The discovery is still preliminary
There is an important limitation to keep in mind. Anthropic has released the research as an early result, and scientists are still investigating what ARTs actually do.
The company has published a preprint describing the work, but more research will be needed before the significance of the system is fully understood. That makes the discovery interesting less because of what ARTs might eventually become and more because of how they were found.
Claude was not simply answering a question with information it already knew. The AI agents searched a large dataset, noticed an unusual pattern, investigated it and produced a candidate that human scientists considered worth testing.
AI could become a new kind of research assistant
Generative AI first became widely known for producing text, images and code. Research projects like this suggest its next major role could involve helping people discover things that were previously hidden inside enormous amounts of data.
That does not mean AI will suddenly conduct science independently. Experiments, validation and expert interpretation remain essential.
But AI systems capable of exploring thousands of possibilities at once could change how scientists decide what to investigate next. Anthropic’s newly discovered ART system is only one early example.
Its real importance may ultimately be what it demonstrates about a broader idea: AI agents may become useful not only for explaining existing knowledge, but also for helping researchers find genuinely new questions to explore.
