Anthropic's Biology Lab Reports CRISPR-Like Enzyme Find

Most AI announcements are about the model itself: a new release, a better benchmark score, a lower price. Anthropic's latest news is different. It is about what the model found in biological data, and about who did the physical lab work that followed.

Only a week after confirming that it runs a wet biology lab in the Bay Area, Anthropic said the lab had already produced what the company considers a significant result. It describes the find as a new enzyme "system" with "properties reminiscent of CRISPR."

What was found: an enzyme system hidden in viral DNA

CRISPR began as a natural immune system in bacteria, which use it to defend themselves against viruses. Researchers later turned it into a gene-editing technology, and it is now widely used in labs.

Anthropic says its new system comes from the DNA of bacteriophages, the viruses that infect bacteria and replicate inside them. According to the company's researchers, the system had not been identified before, and it resembles CRISPR in one important way: it can "perform operations like cutting, copying, and pasting DNA."

That is the claim. Whether the discovery is as large, or as new, as Anthropic suggests is for the wider research community to judge. CEO Dario Amodei has acknowledged that the work builds on earlier research. Writing on X, he noted that a Stanford team had previously discovered a system "that is in some ways similar to the one Claude found."

Why the focus is on Claude, not only the enzyme

Anthropic is not presenting this as an ordinary lab result. The company's main point is that the discovery was made "mostly, though not entirely, by Claude," as Amodei put it.

The timeline is short. The lab was only set up this spring, and Anthropic declined to say exactly how many months it has been running. Finding something notable within a few months would already be quick. But the team says the search itself took Claude just 21 hours of concerted work. In that time, roughly 950 agents combed through data and used about 210 million tokens between them.

For readers following AI agents in research settings, the numbers are the interesting part. This was not a single model answering a question. It was a large, coordinated search running in parallel, then handed over to people for the physical work.

Humans at the bench: how the lab is set up

That handover may be the most important detail in the announcement. Claude did not run the experiments. People did.

Anthropic describes the facility in plain terms. It says the lab looks like a typical molecular biology lab and works only at the lower biosafety levels, BSL-1 and BSL-2. It does not handle pathogens that can infect humans, and all lab work is carried out by human scientists.

The context explains why this matters. The news that Anthropic runs biology experiments arrived shortly after AI leaders, Amodei among them, publicly said that models have become so capable, and potentially so dangerous, that the industry needs to slow down and build safety-testing procedures. Earlier this month, two Anthropic employees also said publicly that AI could pose a risk to human survival. Those concerns sit alongside growing political pressure, including a proposal for a frontier model pause in Washington.

Amodei has named bioterrorism as one of the misuses of AI he fears most. At the same time, he has written that AI could "cure most diseases in 5-10 years." Running a biology lab shows which way Anthropic has come down on that trade-off: it believes the potential gains justify the risks, as long as the setup keeps humans in control of the physical side.

Not the only lab using AI for biology

Anthropic is far from alone here. Stanford researchers have just published a paper on their work combining large language models with CRISPR. A team at UC San Francisco used AI to design enzymes from scratch. Google launched AlphaFold, its first AI-powered biology research tool, back in 2020.

In other words, AI is already part of biological research well beyond the companies that build the models. What sets Anthropic's case apart is that the model developer is also running the lab, and presenting the model as the main driver of the discovery.

What comes next: automation, with conditions

Amodei has not ruled out a lab where Claude does more than search data. "Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren't doing that today," he said.

The wording is careful. "Eventually" and "with appropriate safeguards" leave plenty of room, and "we aren't doing that today" describes the current limit clearly. For now, the division of labour is simple: the model searches and proposes, and human scientists handle the pipettes.

The open questions are practical ones. Will independent researchers confirm that the enzyme system is new and as capable as described? How much credit belongs to Claude, and how much to the scientists and prior studies it built on? And if the lab ever moves toward autonomous experiments, what exactly will those safeguards look like?

Until those answers arrive, the announcement is best read as two stories in one. It is a claim about a possible new gene-editing tool. It is also an early look at how an AI company wants to use its models in the physical world, and where it says it is drawing the line for now.