Claude AI Discovers a CRISPR-Like Enzyme, But Nobody Knows What It Does

An artificial intelligence system has done something no human scientist had ever done: it discovered a brand-new enzyme system hidden inside the DNA of bacteria-infecting viruses. Anthropic’s Claude AI spent about 21 hours autonomously searching through massive genomic databases and found what it calls “array-associated reverse transcriptases,” or ART — a three-part system whose repeating DNA pattern looks strikingly like the famous gene-editing tool CRISPR.

Anthropic’s Claude AI flagged a CRISPR-like enzyme system after about 950 agents spent 21 hours scanning genomic databases.

Background: CRISPR has revolutionized biology since its discovery. The Nobel Prize-winning technology lets scientists cut, copy, and paste DNA with precision, and it has powered breakthroughs in personalized medicine and the treatment of genetic disease. For years, researchers have hunted for the next CRISPR — another programmable biological system that could do something similar. According to Anthropic, Claude may have just found a candidate.

The announcement came on September 23, when Anthropic unveiled its new life sciences research group and a molecular biology wet laboratory in the Bay Area — a lab the company built earlier this year, in part after acquiring the startup Coefficient Bio for about $400 million in April. ART is the lab’s first major result. Anthropic published its findings in a preprint, which has not yet been peer-reviewed, alongside a blog post. The company’s post on X drew about 15.8 million views.

The 21-Hour Hunt

The scale of the search is hard to grasp. Anthropic set roughly 950 autonomous Claude agents loose on databases containing about 1.9 billion protein clusters, with a simple instruction: look for interesting new examples of reverse transcriptases, the enzymes that read a string of RNA and build a complementary strand of DNA.

The agents could write their own code to do the analysis. They first identified roughly 200,000 reverse transcriptases, then filtered that ocean down to about 3,500 candidate systems, then produced 20 detailed reports. One case stood out.

The pivotal moment was almost poetic. One agent was reading the raw DNA sequence next to an odd-looking enzyme and noticed a regular pattern with its own eyes. “I can see by eye a tandem repeat array,” it wrote in its notes, then asked itself: “a CRISPR-like … repeat array?!” It did what a careful scientist would do — counted the repeats, measured their spacing, compared the layout with known systems, and searched the literature for any earlier report — before filing its case for human review.

The whole computational campaign took about 21 hours and processed roughly 210 million tokens — work that Anthropic says would have taken a human expert weeks to months.

An illustration of bacteriophages — viruses that infect bacteria. Claude found the new system, dubbed ART, inside a jumbo phage.

A Familiar Pattern in Unfamiliar DNA

The discovery lives inside a jumbo bacteriophage — a giant virus that infects bacteria. The underlying reverse transcriptase enzyme had actually been identified in earlier studies. What Claude noticed for the first time were the system’s defining features: a long array of evenly spaced, non-coding DNA repeats sitting next to the enzyme, plus an accessory protein of unknown function beside it.

That array is why the comparison with CRISPR is being made. A CRISPR array stores a bank of different RNA guides — and that bank is what makes CRISPR-Cas systems programmable, turning them into gene-editing tools. Anthropic’s first lab experiments show that the ART array is also expressed as a set of distinct short RNAs, which hints that something similar may be happening.

The company is being careful about the claims. It does not yet know what ART does. What it can say is that this combination of features — a reverse transcriptase, a partner protein, and a CRISPR-like repeat array — has only ever been seen together in a handful of other systems, and all of those are programmable biological machines able to cut, copy, or paste DNA.

CEO Dario Amodei, in a post on X, wrote that the discovery was made “mostly, though not entirely, by Claude,” and suggested the molecular machine could “represent a new gene editing mechanism.” He has long argued that AI could compress decades of biological progress into years. He also acknowledged that a Stanford team had earlier described a system “in some ways similar” to this one.

Importantly, all of the bench work was done by human scientists, and Anthropic says its lab operates only at biosafety levels 1 and 2 and does not handle pathogens that infect humans.

What the Experts Say

The most significant outside voice is Feng Zhang, the MIT and Broad Institute scientist who helped pioneer CRISPR gene editing and reviewed Anthropic’s preprint. “This is an exciting example of how AI agents can contribute to biological discovery,” Zhang said in a statement. “The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research.”

Not everyone is swept away. Microbiologist Kevin Blake noted that Anthropic has not shown the enzyme is even active, and said nothing in the data indicates ART could rival CRISPR. Dimitri Perrin, a computer scientist at the Queensland University of Technology, wrote in The Conversation that the work nevertheless represents a major advancement in AI autonomy — the real story being that an AI did science mostly on its own.

On Hacker News, a 486-point discussion thread pushed back on some of the claims, with commenters picking apart what was truly novel and what was known — a reminder that this is a preprint, not a peer-reviewed paper.

CRISPR, the Nobel Prize-winning gene-editing tool, is the model for what the newly found system might one day resemble. (Illustration)

What This Means

Step back from the caveats and the magnitude is clear: this is the most concrete public example yet of AI doing genuine science. Earlier AI biology milestones mostly involved predicting structures — spectacular, but still answering questions humans had posed. Here, the AI asked the questions, ran the search, noticed the pattern, and made the case.

For the biotech industry, ART is now a molecule to watch. If the system turns out to be programmable like CRISPR, it could open a new path for gene editing — and for everything built on top of it, from therapies for genetic diseases to better tools for basic research. But that “if” is doing a lot of work: most novel biological systems turn out to be evolutionary curiosities rather than revolutionary tools.

For the AI industry, the result lands in the middle of an arms race. Anthropic built a wet lab and bought a biotech startup; OpenAI shipped its own life sciences model, GPT-Rosalind, in April. Both companies are betting that the company whose AI makes the next great biological discovery will own the future of medicine. Anthropic is now openly inviting scientists to propose follow-up research on ART.

What Happens Next

Anthropic says experiments to determine what ART actually does are already underway. The preprint will go through peer review, and independent labs will need to reproduce the findings before anyone can talk seriously about applications.

The company is asking the broader scientific community to help. Researchers can submit proposals to study the system, and Anthropic’s own wet lab will keep running experiments — testing whether ART is active, what it acts on, and whether it can be programmed.

For now, the honest summary is the one Anthropic itself offers: Claude found something no human had ever seen, with a shape that resembles one of the most important tools in modern science — and the most exciting part is that nobody yet knows what it does.