One strange repeating pattern in a large set of DNA took nearly 1,000 Claude agents, 21 hours and 210 million tokens to find. Anthropic says the pattern points to a “novel enzyme system” and compares it to the machinery behind Crispr.
That is a big comparison. The actual claim is much smaller.
What Claude found
This is the first result from Anthropic’s newly launched wet lab. The company said Claude “autonomously discovered” an enzyme system that resembles the machinery behind Crispr, the gene-editing tool.
Anthropic said Claude did the search itself, working through a large database of DNA sequences. Its scientists did only two things: they wrote the first prompt and they did the lab work.
The numbers work out like this. Nearly 950 Claude agents used 210 million tokens over 21 hours. One agent then noticed an unusual repeating pattern in the dataset and flagged it for a human to review.
After more analysis and lab testing, Anthropic said, the team identified “a previously uncharacterized enzyme system found in bacteriophages.” Bacteriophages are viruses that infect bacteria.
Nobody knows what it does yet
This is where the story runs out of detail. Anthropic said it is still working out what the enzyme system actually does.
Even so, the company said it considered it “important to share such findings early, both to demonstrate Claude’s capabilities and to give the broader community insight into what we’re working on.”
The first half of that quote explains most of it. Anthropic wants to show what Claude can do, and it is doing that while it gets ready to go public.
There is no clear sign yet that the discovery will be useful in practice, and matching Crispr’s impact on gene editing would be a much bigger achievement. Anthropic has admitted the announcement is early. It also comes as the company tries to bring more scientists into its lab and move into areas such as drug discovery.
Science as the new benchmark
Anthropic is not alone here. AI companies are relying more and more on scientific research to show that their more capable models have real value.
OpenAI has gone further. It has been working through harder and harder math problems while paying little attention to the field’s academic norms and traditions.
So the Crispr comparison deserves some caution. What exists right now is an enzyme system in viruses that infect bacteria. An agent swarm found it and lab work confirmed it, but nobody can yet explain what it does. That is a real result. Until Anthropic can say what the system does, though, it is a lab note with a famous name attached.