Anthropic Says Claude Found a CRISPR-Like System in 21 Hours

Anthropic says Claude has done more than chat and summarize. In a blog post, the company said its model helped uncover a previously unknown enzyme system in bacteriophage DNA that looks a lot like CRISPR. The search took about 950 AI agents and roughly 21 hours. The catch is obvious: the biology is still a mystery.

Anthropic calls the find “array-associated reverse transcriptases,” or ART. The system appears to have three parts: a reverse transcriptase enzyme, a partner gene with an unknown role, and a long stretch of repeating DNA. Those repeats are what caught attention, because CRISPR itself is built around repeated DNA sequences. In the lab, Anthropic says researchers confirmed the array shows up as short, distinct RNA pieces, which is the sort of signal people look for in programmable DNA systems. But nobody has shown that ART actually edits DNA.

The search itself was a very human-AI relay race. People wrote the initial prompt, then Claude agents screened about 1.9 billion protein clusters and narrowed that to roughly 200,000 reverse transcriptases. From there they cut the list to 3,500 candidates, then 20 especially interesting ones, and produced readable reports. One agent eventually spotted the repeating DNA pattern sitting beside one of the enzyme genes. Anthropic says the whole search used about 210 million tokens. All the wet-lab work was still done by humans.

There are reasons to be cautious. The enzyme itself was already known; Claude’s contribution was linking it to the DNA array and the partner protein. Anthropic also says Stanford researchers had already found a system “in some ways similar” to this one. And when the company reran the search ten times, The Next Web reported that not a single run found the same pattern again. Chance clearly had a seat at the table.

Still, the company is pushing hard into biology. Anthropic has been running a molecular biology lab in the Bay Area since the spring, at BSL-1 and BSL-2 levels only, and it says it isn’t handling human pathogens. It has also bought the biotech start-up Coefficient Bio for about $400 million. That fits Dario Amodei’s bigger pitch that A.I. will matter in life sciences. Whether ART becomes a real editing tool or just another intriguing natural oddity, the episode shows why every CRISPR-adjacent find makes scientists sit up straight.