Anthropic says its Claude AI model helped identify a previously unknown enzyme-like system, provisionally named ART, after 950 Claude-powered agents analyzed DNA sequences for 21 hours. The agents filtered over 200,000 genes down to about 20 candidates that human researchers then tested in the lab. The genomic arrangement resembles CRISPR and initial tests suggest ART could be programmable, but the finding is preliminary and reported only as a preprint. Independent peer review and further study are needed to confirm the discovery and assess its implications.
Anthropic Says Claude AI Helped Identify Potential New Gene‑Editing System, Named 'ART'
According to Anthropic, some 950 Claude-powered software agents combed through a database of DNA sequences for 21 hours (SEBASTIEN BOZON)
US artificial intelligence startup Anthropic on Wednesday announced what it describes as the first discovery from its molecular biology lab: a previously unknown enzyme-like system identified by its Claude model that the company says could accelerate basic biological research.
Anthropic says roughly 950 Claude-powered software agents scanned a database of DNA sequences continuously for 21 hours. The system reportedly flagged more than 200,000 genes from a single enzyme family and autonomously narrowed those to about 20 candidates for experimental follow-up — a filtering task the company says would otherwise take an expert researcher weeks or months.
According to Anthropic, human researchers provided the initial instructions and performed lab experiments, while the AI agents autonomously decided which leads to pursue. One agent detected a pattern in bacteriophages (viruses that infect bacteria): short, repeating DNA segments located next to two genes, one encoding a known enzyme and one for an uncharacterized protein.
The arrangement resembles CRISPR, the bacterial defense system that scientists repurposed into a powerful gene-editing tool. Anthropic has provisionally named the newly identified system "ART." The company says the biological function of ART remains unknown, but its initial experiments suggest the system might be programmable in a way analogous to CRISPR.
Dario Amodei, Anthropic CEO: "The finding is preliminary, but it may constitute a new gene-editing mechanism that could have applications in gene therapy." He noted the result while urging caution on AI development in remarks to the UN Security Council.
The results are published as a preprint and have not yet undergone peer review. Anthropic released statements saying that Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, called the repeated DNA fragments "genuinely intriguing" and that the observation "merits further investigation." Anthropic also noted that a separate team at Stanford described a comparable but distinct system.
Safety, Oversight and Next Steps
Anthropic emphasized that its Bay Area lab does not work with pathogens capable of infecting humans and that all experiments so far were carried out by human scientists. While Amodei did not rule out the possibility of Claude operating lab equipment in the future, he said any such steps would require appropriate safeguards.
Independent validation and peer review will be essential to confirm ART's biological role and its potential as a tool. If substantiated, a programmable system analogous to CRISPR could open new research and therapeutic avenues — but it would also raise important safety, ethical and governance questions that the scientific community must address.
Help us improve.
