Anthropic’s Claude finds new CRISPR-like system in viruses

Anthropic says its Claude AI has found a previously unknown enzyme system inside viruses that infect bacteria, and its layout looks a lot like CRISPR, the natural machinery behind the world's best-known gene editing tool. The company has named it array-associated reverse transcriptase, or ART, and says Claude did most of the work after its scientists gave it a single broad prompt. The finding is the first result from Anthropic's new life sciences research group and wet lab, set up in the Bay Area this spring. What ART does is still unknown. CEO Dario Amodei said as much in a post on X, noting that its precise function, biotech usefulness "(if any)" and significance are not yet clear. "But at minimum it is work I would have been proud to do as a PhD student," he wrote. Claude spotted a CRISPR-like pattern that earlier studies missed ART has three parts. There is a reverse transcriptase (RT), an enzyme that copies RNA into DNA. There is a partner gene of unknown function sitting beside it. And there is a long array of evenly spaced DNA repeats. That repeat array is what invites the CRISPR comparison. In CRISPR systems, a similar array stores a bank of RNA sequences, which is what makes them programmable. The RT itself, found in a jumbo phage, had been identified in earlier research. Anthropic says Claude appears to be the first to notice the array and the accessory protein around it. Early lab tests show the array is expressed as a set of distinct short RNAs, which suggests ART may work in a way similar to CRISPR. Experiments to confirm its function are ongoing, and a pre-print is already out. Around 950 Claude agents searched DNA data for 21 hours Anthropic says roughly 950 Claude agents spent 21 hours combing through a massive database of DNA sequences, using 210 million tokens along the way. They gathered more than 200,000 RTs, shortlisted 3,500 new candidate systems and narrowed those to 20, each with a report written for human review. The company says this kind of analysis can take an expert scientist weeks to months. The moment of discovery came while one agent was reading raw DNA next to an odd-looking RT. Its note read: "that's a CRISPR-like … repeat array?!" It then counted the repeats, measured their spacing, compared the layout with known RT systems and checked the literature before flagging it. Amodei described the split carefully. The work was done "mostly, though not entirely, by Claude," he wrote. The team picked the research area, Claude read the literature and genome data and proposed experiments, and human scientists ran them. The lab only handles biosafety level 1 and 2 work, with no pathogens that infect people, and the team used Claude Science and Claude Code, the same tools any scientist can access. CRISPR pioneer Feng Zhang calls the finding worth a closer look Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute, reviewed the pre-print. He called it "an exciting example of how AI agents can contribute to biological discovery" and said the RNA-repeat arrays were "intriguing and merits further investigation." Amodei also pointed out that a Stanford team, working separately, recently described a different RT system with a similar array. The two evolved independently, he said. Dario Amodei says AI in biology is on the same curve as maths Amodei used the announcement to make a wider case. In 2023, he wrote, AI models struggled with high-school maths. By late 2026, they are beginning to solve some of the hardest open problems in the field. He believes biology is on a similar exponential trend. Biology, of course, needs lab work. Amodei says humans working with AI can validate key results in a few weeks, and that Claude may one day run lab equipment itself with safeguards in place, though Anthropic isn't doing that yet. He was clear that faster discovery won't shorten clinical trials. It could, however, send many more promising drug candidates into the pipeline. That links back to his 2024 essay Machines of Loving Grace, where he wrote that AI could help "cure most diseases in 5-10 years." Anthropic now plans to expand its life sciences team and is inviting outside scientists to pitch collaborations. Whether ART ever becomes a tool on the scale of CRISPR is anyone's guess. What Anthropic has shown is that Claude can find something in public DNA data that had been sitting there, unnoticed, until now.