Anthropic’s Claude AI Identifies New Enzyme System in Bacteria-Infecting Viruses

Anthropic says its Claude AI has helped identify a previously unknown enzyme system in viruses that infect bacteria, uncovering a biological structure that appears to share similarities with CRISPR. The system, which Anthropic has named array-associated reverse transcriptase (ART), was discovered through an AI-assisted analysis of large amounts of publicly available DNA sequence data.

The company says Claude identified a previously overlooked combination of a reverse transcriptase, an adjacent accessory gene and a long array of repeated DNA sequences. Researchers are now conducting laboratory experiments to determine what the system actually does and whether it could have any future biotechnology applications.

Anthropic CEO Dario Amodei said the discovery was produced mostly, although not entirely, by Claude. He also cautioned that the biological function and potential usefulness of ART remain unknown.

How Claude Identified the CRISPR-Like System

ART contains three notable components: a reverse transcriptase, a nearby gene whose function has not yet been established, and a long sequence of evenly spaced DNA repeats.

The reverse transcriptase itself had already been identified in previous research. According to Anthropic, Claude's contribution was recognizing the significance of the repeat array and the surrounding accessory protein.

The repeated DNA structure prompted comparisons with CRISPR. CRISPR systems use arrays of genetic sequences that can produce RNA molecules involved in biological targeting and defence mechanisms.

Early laboratory experiments reported by Anthropic found that the ART repeat array is expressed as a collection of distinct short RNA molecules. Researchers say this could indicate a functional relationship with CRISPR-like systems, although further experiments are needed to establish exactly how ART works.

A preprint describing the research has also been released.

Nearly 950 Claude Agents Analysed DNA Sequences

Anthropic says approximately 950 Claude agents worked on the analysis for about 21 hours, processing around 210 million tokens.

During the research, the AI systems reportedly collected more than 200,000 reverse transcriptases, identified around 3,500 candidate systems, and eventually narrowed the list to 20 systems for human researchers to examine.

Anthropic says this type of analysis can require scientists to spend weeks or months manually examining biological data.

The potential ART discovery emerged when one AI agent was analysing DNA surrounding an unusual reverse transcriptase. The agent identified a repeated sequence pattern and investigated whether it resembled the architecture of known CRISPR systems.

Claude then examined the repeat structure, compared its spacing with other systems, reviewed relevant scientific literature and flagged the finding for further investigation.

Human Scientists Still Conducted the Laboratory Work

Despite the extensive role attributed to Claude, Anthropic says human researchers remained responsible for the experimental work.

The team selected the research direction, while Claude analysed scientific literature and genomic data and proposed experiments. Human scientists then conducted the laboratory experiments.

Anthropic also said its laboratory work is limited to biosafety level 1 and 2 research and does not involve pathogens that infect humans.

The research used Anthropic's Claude Science and Claude Code tools, according to the company.

CRISPR Pioneer Feng Zhang Reviews the Finding

Feng Zhang, a CRISPR researcher and professor at MIT and the Broad Institute, reviewed the preprint.

Zhang described the research as an example of how AI agents could contribute to biological discovery. He also said the RNA-repeat arrays identified in the research warrant further investigation.

Anthropic noted that researchers at Stanford have separately reported another reverse-transcriptase system with a similar array. Amodei said the two systems appear to have evolved independently.

Dario Amodei Sees a Growing Role for AI in Biology

The discovery has also prompted Amodei to make a broader argument about the potential development of AI-assisted scientific research.

Amodei compared the progress of AI in biology with advances in mathematical problem-solving. He argued that increasingly capable AI systems could eventually help researchers investigate biological questions at a much larger scale.

However, biology still requires physical laboratory experiments to validate computational findings.

Amodei suggested that AI could eventually interact directly with laboratory equipment under appropriate safeguards, although Anthropic is not currently operating its systems in that way.

He also acknowledged that faster scientific discovery would not eliminate the need for clinical trials. Instead, he argued that AI could potentially increase the number of promising drug candidates entering the development pipeline.

Could ART Become a New Biotechnology Tool?

It is still too early to determine whether ART could become a useful biotechnology platform.

Anthropic has not established the precise biological function of the system, and the company itself says its potential usefulness remains uncertain. The current research demonstrates that AI can help identify previously overlooked patterns in large biological datasets, but laboratory experiments will be necessary to determine whether those patterns translate into a practical biological mechanism.

Anthropic plans to expand its life sciences research efforts and is inviting scientists to collaborate with its team.

For now, the significance of ART remains an open scientific question. The discovery nevertheless provides an example of how AI systems can be used to search enormous biological datasets and identify relationships that may warrant further human investigation.

Read More : Anthropic Combines Claude Chat and Cowork Into One Interface for Paid Users