When AI Found CRISPR's Cousin: Claude's Autonomous Discovery of a Novel Enzyme System

In the spring of 2026, Anthropic formed a life sciences research group with a radical question: can a general-purpose AI model systematically discover new biology? It was a question that bordered on audacious. AI models had been used to predict protein structures, design drug candidates, and analyze genomic data — but discovering entirely new biological systems? That was the domain of human scientists, not machines. The answer arrived faster than anyone expected.

After 21 hours of searching through massive DNA databases, roughly 950 Claude agents using 210 million tokens found something no human scientist had ever noticed — a novel enzyme system with properties eerily reminiscent of CRISPR. The discovery has sent ripples through the molecular biology community, and it represents what may be the first instance of an AI system autonomously discovering a novel biological mechanism through large-scale data analysis.

The Discovery That Shouldn't Have Been Possible

The system Claude discovered is called Array-Associated Reverse Transcriptases (ART). It consists of three parts: a reverse transcriptase (RT) enzyme that copies RNA into DNA, a partner gene beside it, and a long array of evenly spaced DNA repeat sequences. That repeat layout looks astonishingly like a CRISPR array — the same kind of programmable DNA-targeting system that launched the gene-editing revolution and earned a Nobel Prize in Chemistry in 2020.

What makes this discovery extraordinary isn't just the biology. It's how it happened. Claude received only a high-level prompt to search through DNA sequence databases for interesting new examples of reverse transcriptases. From there, the agents operated autonomously through a multi-stage pipeline:

For a single human scientist, this type of analysis would take weeks to months of painstaking work — downloading sequences, running alignment tools, inspecting genomic neighborhoods one by one, and cross-referencing findings against the literature. Claude did it in under a day, and it did so with a level of thoroughness that would be impossible for a human working alone. The 950 agents worked in parallel, each examining different branches of the protein family tree, sharing findings, and building on each other's discoveries.

The Moment of Discovery

The most remarkable part of the story is the moment Claude found it. While combing through raw DNA sequences near an unusual reverse transcriptase, one of the agents essentially exclaimed:

"[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array ... that's a CRISPR-like ... repeat array?!"

It then did exactly what a human scientist would do upon making an unexpected observation. It counted the repeats, measured their spacing, compared the layout with known RT systems, and searched the literature for any previous report of the pattern. After thorough analysis, it was convinced it had found something genuinely new — and filed a report for human review. The moment of discovery — an AI agent recognizing a pattern and getting excited about it — is a glimpse of something genuinely novel in the history of science. The agent didn't just flag a statistical anomaly; it understood what it was looking at well enough to recognize its significance.

Why ART Matters

The systems that share ART's characteristics — a reverse transcriptase paired with a DNA repeat array — are extraordinarily rare. The handful of other known systems with these properties are all programmable and perform operations like cutting, copying, and pasting DNA. CRISPR-Cas9, which transformed science and medicine, is the most famous example. If ART proves to be similarly programmable, it could open up entirely new approaches to gene editing and biotechnology — potentially a new class of tools for precise DNA manipulation that work through different mechanisms than CRISPR.

Early experiments confirm that the ART array is expressed as a set of distinct short RNAs — the same pattern that makes CRISPR programmable. In CRISPR systems, these short RNAs (called guide RNAs) direct the Cas9 enzyme to specific DNA sequences for cutting. If ART's short RNAs similarly direct the reverse transcriptase to specific sequences, the system could be repurposed for targeted DNA writing — potentially more versatile than CRISPR's cutting-based approach. This suggests something analogous may be at play, though the primary function of ART is still under investigation.

The biotechnology implications are significant. CRISPR revolutionized gene editing by making it programmable — scientists could target any DNA sequence by simply changing the guide RNA. If ART is similarly programmable, it could enable new types of genetic modifications that CRISPR cannot easily achieve. Reverse transcriptases copy RNA into DNA, which means ART could potentially be used to insert new genetic information at targeted sites, rather than just cutting existing DNA. This would be a fundamentally different capability with profound implications for gene therapy, synthetic biology, and basic research.

A New Way of Doing Science

This discovery signals something bigger than a single enzyme system. It represents a fundamentally new way of conducting biological research, where AI agents collaborate with humans at every step of the scientific process:

The scale is staggering. A single research campaign can produce hundreds to thousands of candidate reports. Anthropic's team has started studying the hypotheses themselves as an object of research — asking what distinguishes proposals worth testing from those set aside, and feeding those insights back into Claude's instructions to develop what amounts to scientific taste. This meta-level analysis — studying how the AI does science in order to improve its scientific reasoning — is itself a novel research approach.

The Bigger Picture

Many of the discoveries that revolutionized biology started with a scientist noticing something odd in nature's molecular machines. Restriction enzymes were found in bacterial immune systems and launched the biotechnology industry. Taq polymerase came from a Yellowstone hot spring bacterium and became the basis for PCR, the technique that powers virtually all molecular biology. CRISPR was first noticed as unusual repeat sequences in bacterial DNA and is now the foundation of gene-editing medicines that have already treated patients with sickle cell disease.

ART might be the next entry in that list — and it was found not by a human staring at a gel or a sequence alignment, but by an AI agent reading raw DNA and getting excited about a repeating pattern it recognized. The fact that Claude found this in a jumbo phage reverse transcriptase that had been studied before — but whose defining features were never noticed — demonstrates exactly what AI-driven discovery brings to the table: the ability to process volumes of data that no human could manually review, while applying reasoning that goes beyond simple pattern matching to recognize the biological significance of what it finds.

What Comes Next

Anthropic has released a pre-print with full technical details and is actively working to understand ART's primary function. The company is also inviting external scientists to propose research questions, extending this AI-driven approach to a broader range of problems in genomics and beyond. The open invitation is notable — it suggests Anthropic views this not as a proprietary capability but as a new scientific method that should be shared with the research community.

Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute, reviewed the pre-print and said: "This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation." When a CRISPR pioneer says your discovery merits investigation, that is not faint praise.

When the tool that discovered CRISPR's successor is not a microscope or a sequencer but a language model, something fundamental has shifted in how science gets done. The question is no longer whether AI can make scientific discoveries — Claude has answered that question definitively. The question is how many more discoveries are hiding in plain sight, waiting for an agent to notice them. The DNA databases are growing exponentially. The literature is expanding faster than any human can read. And the AI models are getting better every month. We may be at the beginning of a new era of scientific discovery, where AI agents and human researchers work together to uncover the secrets of biology at a pace that neither could achieve alone.