**Anthropic**-developed **Claude** discovered a previously unreported new enzyme system in bacteriophage (a virus that infects bacteria) DNA. In other words, AI picked up structural features that prior research had overlooked.
Core content
After roughly 950 Claude agents searched a large DNA database and identified 3,500 candidate systems, they were narrowed down to the final 20.
The newly characterized ART system consists of a reverse transcriptase, a partner protein whose identity is still unknown, and arrays of repetitive sequence elements.
The research team has not yet identified ART’s biological function, nor is there evidence that it can be used as a gene-editing tool.
A new enzyme system discovered by Claude
On September 23, Anthropic revealed results from running about 950 Claude agents with a single prompt to find a new reverse transcriptase in a large DNA database. In the process, the agents used a total of 210 million tokens over 21 hours, collected more than 200,000 reverse transcriptases, and narrowed them down to 3,500 candidate systems.
After that, human researchers selected 20 final candidates for review, and Anthropic explained that if human experts performed this level of analysis directly, it could take anywhere from weeks to months.
In particular, a standout candidate was a case where, alongside a gene for a reverse transcriptase, there was an unrecognized partner protein and a DNA repeat array with regular spacing reminiscent of a CRISPR array.
Anthropic named this ART—short for “array-associated reverse transcriptases” (“array-associated reverse transcriptases,” array-associated reverse transcriptases). The lab researchers also confirmed that the repeat arrays of ART produce different short RNAs.
Related article: Ethereum retries with a $3,000 push, powered by $30,000 worth of open interest
Anthropic's bio strategy and its significance
Although ART’s structural resemblance to CRISPR has drawn interest in academia, researchers caution against viewing it immediately as a “second CRISPR.”
MIT and the Broad Institute CRISPR pioneer Feng Zhang reviewed the preprint paper that has not yet undergone peer review and said the RNA repeat array is “very interesting and worth further study.”
Meanwhile, Kevin Blake, a microbiologist at the University of Washington School of Medicine, takes a more cautious view. He points out that there was a wide gap between nature’s CRISPR systems and the gene-editing technologies developed by scientists, and he believes it’s hard to expect ART to immediately translate into a practical technology.
Anthropic also said there is no direct evidence yet that ART can cut, copy, or paste DNA, and experiments to verify its actual function are ongoing.
Anthropic CEO Dario Amodei says ART may be possible as a new kind of gene-editing mechanism, but he emphasizes that it’s still too early to judge its usefulness and scientific or commercial significance.
He has argued that recent breakthroughs in ultra-fast AI could greatly accelerate the pace of discoveries in biotechnology, and he has also offered the outlook that AI could contribute to developing most treatments for diseases within the next 5 to 10 years. This ART case is drawing attention as the latest example testing the possibility of “AI-driven bio innovation.”
Next read: If AI investment doesn't deliver expected productivity, a $1 trillion-scale adjustment could come – MIT