Claude found a CRISPR-like puzzle hidden in phage DNA

Anthropic says Claude has found a DNA system nobody had described before, and nobody yet knows what it does: a gene for an unknown enzyme in bacteriophages, with a long array of repeating DNA right beside it that looks somewhat like CRISPR. According to Anthropic's post on X, it is the first result from the company's new molecular biology lab.

At a glance

• Claude sifted through data and scientific literature to propose hypotheses and candidate systems, and Anthropic's biologists reviewed them, tested the most promising ideas and did all of the lab work themselves.
• Only a handful of known systems share its features, and all of them can cut, copy and paste DNA; CRISPR, a programmable system of that kind, now underpins genetic medicines.
• The catch is the obvious one: Anthropic says it does not yet know what the system does, and working that out, and whether it has similar uses, will take much more work.

If you have not been following, Anthropic has spent months building up its life sciences work. In April 2026 it paid about $400 million in stock for Coefficient Bio, a startup whose middleware lets AI models control molecular design software, run sequence alignments and query structural databases directly, and it added Novartis CEO Vas Narasimhan to its board.

Claude proposed the candidate, and Anthropic's biologists ran every experiment

The division of labour is spelled out carefully. Claude works through data and literature and generates hypotheses and candidate biological systems worth studying. Anthropic's scientists review those hypotheses, pick the most promising ones and test them, and all of the lab work is done by the scientists, not by Claude.

The phage system is the first result from this setup, which Anthropic describes as a new molecular biology lab where a team of its biologists uses Claude to explore and accelerate fundamental biology research. The company says it wants to extend the approach to a broad range of problems, in genomics and in other fields, and it is asking researchers to send in proposals for research questions.

The new system pairs an unknown enzyme gene with a CRISPR-like repeat array

Bacteriophages are the viruses that infect bacteria, and Claude found the new system hidden in their DNA. It has two visible parts: a gene for a previously unknown enzyme and, sitting right beside that gene, a long array of repeating DNA. A gene parked next to a repeat array is also the familiar layout of CRISPR loci, which is where the comparison comes from.

The company the system keeps is where it gets interesting. Only a handful of known systems share its features, and all of them are able to cut, copy and paste DNA. Anthropic notes that finding programmable systems like these has historically helped revolutionize medicine; CRISPR, for instance, is now the foundation of genetic medicines.

How do CRISPR-like systems cut or move DNA?

Classic CRISPR is a search-and-cut tool. A bacterium stores short copies of old invaders' DNA between the repeats of its array, turns them into guide RNAs, and a Cas enzyme uses those guides to find and cut matching DNA. Think of the array as a board of wanted posters and the enzyme as the officer who checks every stranger against it.

Some relatives do not cut at all. According to Wikipedia, CRISPR-associated transposons, or CASTs, are mobile genetic elements that use minimal CRISPR systems for RNA-guided movement of their own DNA, and they lack the proteins or domains that cleave DNA. Instead, transposon machinery similar to the Tn7 transposon teams up with CRISPR RNA and Cas proteins, moving DNA through a two-part transposase, TnsA and TnsB, and a regulator protein, TnsC.

Anthropic's post does not name the handful of known systems its find resembles, so it does not say whether the new enzyme looks more like a cutter, a copier or a mover. The repeat array only tells you where to look.

Anthropic has also built a physical wet lab in the San Francisco Bay Area

Separately from the phage result, Anthropic has quietly assembled a physical wet lab in the San Francisco Bay Area, where it wants Claude to direct robots through real experiments. Eric Kauderer-Abrams, the company's head of life sciences, confirmed the lab to Reuters.

We believe that to do biology, the final test is still and will be for a while in real lab work.

He also said the company is "in the very early innings of using AI to automate the execution of lab work," and that life sciences is already one of Anthropic's largest areas of investment by personnel and resources. Humans still supervise the lab for safety, no clinical trials are attached to the work, and a spokesperson told Reuters the lab is not specifically for drug discovery.

According to Startup Fortune, Anthropic launched Claude Science, an AI workbench and in-house drug discovery push, at a San Francisco event on June 30, 2026. The same outlet reports that in August 2026 Claude-designed protein binders succeeded against 14 of 15 targets tested by Adaptyv Bio and Twist Bioscience, with success rates from 22.6% to 35.1% depending on the model and setup.

The post leaves out most of what a biologist would ask first: which phages carry the system, how many copies turned up, and what exactly the lab tested. We know the scientists tested the most promising ideas, but not whether this candidate has been checked at the bench or only mapped in sequence data. In our view, the careful wording, an enzyme system that "looks somewhat similar to CRISPR" with an unknown function, is the right level of claim for what has been shown.

What the repeat array must prove

The next step is function. Anthropic says it will take much more work to learn what the system does and whether it can be put to uses like CRISPR's, and it gives no timeline for that work or for a paper. In the meantime, the company is collecting research-question proposals for the same Claude-plus-biologists approach, in genomics and in other fields.