Claude's enzyme find began with five examples from scientists

According to The New York Times, Claude's hunt started with a small handout: Anthropic's scientists gave the AI five examples of reverse transcriptases, the enzymes that copy RNA into DNA. The search that followed produced the company's first biology discovery, a novel CRISPR-like system, and the Times asks how much of that discovery the AI can claim as its own.

At a glance

• Anthropic's biology lab has announced a novel CRISPR-like system as its first discovery. Phys.org describes it as a previously unknown enzyme system spotted by the company's Claude model.
• By Anthropic's own account, its scientists supplied only the initial prompt and the lab work. Claude agents combed the database, investigated distinct reverse transcriptase families and picked candidates themselves.
• TechCrunch reports that humans are still, so far, in the loop in Anthropic's biology lab. People also ran the wet-lab experiments, so the agents handled the database search and little else.

Anthropic formed its life sciences research group in the spring of 2026 to test whether general AI models can systematize and speed up biological discovery. That meant building its own lab and a single team covering everything from training Claude in biology to running experiments. The Scientist describes the project as a new wet lab for overlooked biology problems. According to Unite.AI, the team's earlier work covered CRISPR evolution, enzymes for gene therapies and tools for spotting anomalies in DNA.

Anthropic's scientists wrote the prompt and ran the lab, and Claude agents did the search

The Times account starts with the five example reverse transcriptases, or RTs. From there, the paper reports, the agents wrote their own software to search databases for similar genes and look at them closely. Anthropic says the human role stopped at the initial prompt and the lab work. The agents combed the database, worked through distinct RT families and used their own judgment to choose candidates.

The scale figures come from Unite.AI, which reports Anthropic's numbers. Roughly 950 agents searched the data for 21 hours and used 210 million tokens before one of them flagged a repeating pattern of DNA sequences next to the gene for an odd-looking RT. Across the whole campaign the agents gathered over 200,000 RTs, picked out 3,500 new candidate systems and narrowed them to the 20 most compelling.

Anthropic's lab calls its first discovery a novel CRISPR-like system

According to Unite.AI, Anthropic named the system array-associated reverse transcriptases, or ART, and found it mainly in bacteriophages, the viruses that infect bacteria. Anthropic's own write-up says the RT at its centre, found in a jumbo phage, had already been identified in earlier studies. The new part, it says, is what sits around that enzyme: an array of non-coding DNA sequences and an extra accessory protein of unknown function.

Unite.AI also describes how the find happened. While reading the raw DNA next to the RT, the agent wrote that the sequence was "spectacular" and that it could see a tandem repeat array by eye. It then counted the repeats, measured their spacing, compared the layout with known RT systems and checked the literature for earlier reports. After that it filed a report for human review.

Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute, reviewed the preprint. According to Anthropic, he called the RNA-repeat arrays next to reverse transcriptases "genuinely intriguing" and said they merit further investigation.

Why a repeat array next to a copying enzyme stood out

A reverse transcriptase reads an RNA molecule and writes a matching DNA copy. According to Wikipedia, HIV and hepatitis B use one to replicate their genomes, and retrotransposons use one to spread through a host genome. Eukaryotic cells use one to extend telomeres. A CRISPR array is a row of short repeated DNA sequences separated by unique stretches. In bacterial immune systems, those stretches work as a record of past infections.

So a row of repeats sitting right beside an RT gene looks CRISPR-like. Think of finding a filing cabinet bolted to a photocopier. You still don't know what gets filed, but the pairing deserves a closer look. Anthropic puts the find in a line of breakthroughs that began with someone noticing something odd: restriction enzymes, Taq polymerase from a Yellowstone hot spring, and CRISPR itself, first spotted as an unusual repeat.

The biology is still open. Anthropic says the accessory protein's function is unknown, and none of the accounts explains what ART actually does in a cell. In our view, "on its own" fits the 21-hour search that Unite.AI describes much better than the discovery as a whole. Humans chose the five starting enzymes, ran the lab work and reviewed what the agents reported.

What ART still has to prove

So far the work is a preprint, and even Zhang's praise ends with a call for more investigation. No date has been given for peer review. No timeline has been given for experiments that would show what the arrays and the mystery protein do. The open question is whether the same agent setup can make a second find with less human help at the start.