Claude’s enzyme find began with five examples from humans

Claude’s search started from five reverse transcriptases, enzymes that copy RNA into DNA, which Anthropic scientists handed it as examples, The New York Times reported on September 27. Starting from those five, the AI agents wrote their own software to comb databases for similar genes. The result was a CRISPR-like system that Anthropic’s new biology lab presented as its first discovery. The Times asks whether a find that began that way counts as one Claude made on its own.

Anthropic tells it more simply. In its own post, the company says it gave Claude a prompt to search a massive database of DNA sequences for interesting new reverse transcriptases, and that its involvement was limited to that initial prompt and the lab work. That description has to stretch to cover the five examples the Times reports and a team of biologists running the experiments. TechCrunch reported that humans stayed in the loop, and that Anthropic did not let Claude run loose in the lab unsupervised.

The search was large. A swarm of around 950 specialised Claude agents went through more than 200,000 known reverse transcriptases in roughly 21 hours, using about 210 million tokens. From about 3,500 candidate systems the AI picked 20 it considered particularly promising and wrote detailed reports on each, Business Standard reported. The outlet says the process was meant to copy a researcher’s routine: find something that does not fit known patterns, investigate it, and check whether anyone has described it already.

The find itself came from one agent reading raw DNA sequences. Business Standard says it flagged an unusual pattern, recognised that it resembled a CRISPR-style repeat array, counted the repeats, measured the gaps between them and then searched the scientific literature.

The enzyme was not new. The reverse transcriptase at the centre of the system sits in a jumbo phage, a virus with an unusually large genome that infects bacteria, and earlier studies had already identified it. Anthropic says Claude appears to be the first to notice the features that define the system: an array of non-coding DNA sequences next to the gene, and an extra accessory protein of unknown function. The company calls the system ART, for array-associated reverse transcriptase. It pairs the enzyme with that partner gene and an evenly spaced repeat array. Initial experiments showed the array is transcribed into distinct short RNAs. What the system does is still unknown, and the paper is a pre-print that has not been peer reviewed.

Business Standard also reported that some scientists were quick to cast doubt on the achievement, without saying which part of the work they question. The best-known outside voice on record is supportive. Feng Zhang of MIT and the Broad Institute reviewed the pre-print and called it an exciting example of how AI agents can contribute to biological discovery. He said RNA-repeat arrays tied to reverse transcriptases are genuinely intriguing and merit further investigation.

The lab behind all this is young. Anthropic says it formed the research group in the spring of 2026 to test whether general AI models can make discoveries like restriction enzymes, Taq polymerase and CRISPR more systematic and faster. It built its own lab so that a single team covers everything from training Claude in biology to running experiments at the bench.

Anthropic’s own head of life sciences is careful about the size of the claim. Eric Kauderer-Abrams said Anthropic is not claiming the discovery has the status of PCR or CRISPR, Scientific American reported. He called it the first discovery in the lab that Claude made on its own, said the work is very early, and said Claude is no replacement for biologists.