Claude Finds a CRISPR-Like Enzyme System Hidden in DNA

Key Points

• Anthropic says Claude identified a previously uncharacterized enzyme system in bacteriophages.
• The system, called array-associated reverse transcriptases, or ART, has DNA repeats that resemble CRISPR arrays.
• Around 950 Claude agents searched 210 million tokens of DNA-related data over 21 hours.
• The AI narrowed more than 200,000 reverse transcriptases down to 20 promising candidates for closer analysis.
• Scientists are still testing ART to understand what it actually does and whether it could eventually have a useful biological application.

Anthropic is taking Claude into the biology lab, and one of its first research projects has produced an unusual result. The company says Claude helped identify a previously uncharacterized enzyme system hidden in DNA, one that contains a pattern of repeated sequences reminiscent of the systems behind CRISPR gene editing.

Anthropic has launched a new life sciences research group focused on using AI to examine enormous collections of DNA sequences, generate biological hypotheses and identify unusual proteins for laboratory testing. The company says the work is designed around AI agents and human scientists working together rather than having AI operate independently in the lab.

The discovery began with a massive search for reverse transcriptases, or RTs. These are enzymes that can copy RNA into DNA and are found in a wide range of organisms and viruses. Claude agents gathered more than 200,000 RTs and identified around 3,500 candidate systems before narrowing the list to 20 that appeared particularly interesting.

The search involved roughly 950 Claude agents working for 21 hours and processing about 210 million tokens. During the analysis, one agent noticed something unusual next to a particular RT gene. The DNA contained a long series of repeated sequences that looked surprisingly similar to the repeat arrays associated with CRISPR systems.

That observation led the researchers to investigate the surrounding DNA more closely. They found a system consisting of three main components, a reverse transcriptase, a neighboring partner gene and a long array of evenly spaced DNA repeats. Anthropic has named the system array-associated reverse transcriptases, or ART.

The comparison with CRISPR is interesting because CRISPR systems also contain repeated DNA sequences. Those arrays can produce short RNA molecules that help CRISPR-associated proteins identify specific genetic material. Anthropic's early experiments found that the ART repeat array is also expressed as a collection of short RNAs, although researchers do not yet know whether they perform a similar role.

That uncertainty is important. Anthropic is describing ART as a new biological system, but its exact function has not yet been established. The company says further laboratory experiments are underway to determine how the system works and what the different components actually do.

The discovery also shows a different way AI could contribute to biological research. Much of genomic research involves searching enormous sequence databases for patterns that may be difficult for humans to spot manually. Anthropic says an expert scientist could spend weeks or months carrying out the kind of analysis Claude performed during this project.

Feng Zhang, a CRISPR genome-editing pioneer and professor 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.

Anthropic says the laboratory work itself is performed by human scientists. Its Bay Area lab operates at BSL-1 and BSL-2 levels and does not handle pathogens capable of infecting humans.

For now, ART is better viewed as an early biological finding rather than a new gene-editing technology. The researchers are still trying to understand its basic function. But the project offers an interesting glimpse at how AI could help scientists search through massive biological datasets, spot unusual patterns and turn those observations into experiments that humans can test in the lab.