Claude Finds an Unknown Biological System
Anthropic says Claude has discovered a previously uncharacterized enzyme system hidden in the DNA of bacteriophages, the viruses that infect bacteria. The system contains a reverse transcriptase alongside a long array of repeating DNA sequences that resembles the repeat structure associated with CRISPR. Anthropic has named the newly identified system array-associated reverse transcriptases, or ART.
The discovery is significant because the system’s actual biological function remains unknown. Anthropic says it has not yet established whether ART can perform gene editing or whether it has any practical biotechnology application. The company is therefore presenting the finding as an early biological discovery rather than a new gene-editing technology.
Claude’s role went beyond summarizing existing research. Anthropic said its life sciences team gave the system a broad research direction and allowed Claude agents to examine scientific literature and large collections of DNA sequences. Human researchers then reviewed the resulting hypotheses and carried out laboratory experiments to investigate the most promising candidate.
According to Anthropic, roughly 950 Claude agents spent 21 hours analyzing more than 200,000 reverse transcriptases and identified about 3,500 candidate systems. Those candidates were narrowed to 20 that received deeper analysis before the unusual reverse transcriptase and its neighboring DNA repeat structure attracted particular attention.
The underlying reverse transcriptase was not itself unknown. Researchers had previously identified the enzyme family in a jumbo bacteriophage. What Claude appears to have recognized was the combination of the reverse transcriptase with an adjacent accessory gene and a long array of evenly spaced, non-coding DNA repeats. Anthropic says that combination had not previously been characterized as a biological system.
From AI Hypothesis to Laboratory Validation
The comparison with CRISPR comes from the structure of the DNA repeats, not from evidence that ART performs the same function. Anthropic’s first experiments found that the repeat array is expressed as distinct short RNAs, which the researchers say could point to an important role for the repeated sequences. Further experiments are still underway to determine what the system actually does.
The discovery has attracted attention partly because of who was involved in reviewing it. Feng Zhang, a CRISPR genome-editing pioneer at MIT and the Broad Institute, described the identification of RNA-repeat arrays associated with reverse transcriptases as intriguing and said it merits further investigation. His comments support the scientific interest in the finding, but they do not establish ART as a functional gene-editing tool.
Anthropic sees the project as an example of a broader research model in which AI systems search enormous biological datasets, generate hypotheses and help scientists decide which discoveries deserve experimental testing. The company established a dedicated life sciences research group and laboratory in 2026 specifically to investigate this approach.
The model is particularly relevant to genomics because the volume of available sequence data has grown far beyond what individual researchers can manually inspect. Genome mining already plays an important role in discovering unusual enzymes and biological systems, but AI agents can potentially examine much larger numbers of candidates and identify relationships that might otherwise remain buried in databases.
Related: Pioneer Labs Creates First Engineered Microbe for Mars as Terraforming Plan Advances
Anthropic argues that this could eventually affect more than fundamental biology. Discoveries of new molecular systems can provide researchers with new tools, possible drug targets and new approaches to manipulating or measuring biological processes. But the company also acknowledges that accelerating discovery does not automatically shorten clinical development, since laboratory validation, safety testing and clinical trials still impose their own timelines.
The ART finding also arrives during a broader expansion of AI-assisted biological research. Stanford researchers recently used an AI model called Evo 2 to design bacteriophages that were subsequently synthesized and tested against E. coli, while Anthropic has been developing specialized models and safeguards for life sciences work. These projects differ substantially, but together they illustrate how AI is moving from analyzing biological information toward participating more directly in scientific discovery.
Anthropic’s claim that AI performance in biology could follow an accelerating trajectory remains a prediction rather than an established scientific law. Biology also differs from mathematics because promising computational hypotheses must ultimately survive experiments conducted in the physical world. In the ART project, humans remain responsible for the laboratory work, and Anthropic says its facility operates at BSL-1 and BSL-2 levels and does not handle pathogens capable of infecting humans.
The next stage will determine whether ART becomes more than an interesting genomic discovery. Researchers need to establish the system’s natural function, understand how its components interact and determine whether those properties have any useful application. For now, Claude has demonstrated something narrower but important: an AI system can search large biological datasets, identify an unusual molecular pattern that researchers had not characterized as a system, propose a hypothesis and help direct human experiments toward investigating it.
About The Author
