Claude AI finds novel viral enzyme system with unknown function

Anthropic's AI model Claude has identified a previously unknown enzyme system in the DNA of bacteriophages. The core enzyme had been recorded in earlier research, but the surrounding components appear to have been missed.

Anthropic published the results on September 23, presenting them as evidence that AI can now identify patterns that human experts failed to see. What the discovery means in practice is not yet clear.

Why Should Anyone Outside a Lab Care?

Many medical breakthroughs originated from a scientist spotting something unusual in nature. For instance, restriction enzymes found in bacteria launched the biotechnology industry.

The enzyme from a Yellowstone hot spring became the basis of PCR. That DNA-copying technique now supports much of modern medical diagnostics.

Claude took that noticing step with only a broad instruction from Anthropic's scientists, the company says. CEO Dario Amodei argues this is part of a larger pattern. In 2023, AI systems struggled with high-school mathematics.

By late 2026, he claims, they are starting to solve some of the field's most difficult open problems. He believes AI for biology is on a similar trajectory.

If Amodei is correct, faster discovery could uncover new drug targets and novel treatments. It would not shorten clinical trials, but it could fill the drug pipeline with more promising candidates.

He has previously written that AI could help cure most diseases within five to ten years, describing that goal as just barely attainable.

For now, human scientists still run every experiment, and the new system's function remains unknown.

What Exactly Is CRISPR?

Bacteria are constantly attacked by viruses called bacteriophages, or phages. Many bacteria defend themselves using CRISPR.

CRISPR acts as a memory bank. Bacteria store short DNA snippets from past invaders between repeated sequences, a stretch scientists call an array.

Each snippet is copied into a short RNA molecule. That RNA then guides a protein to locate and cut matching viral DNA if the same virus returns.

Scientists noticed CRISPR as an unusual repeating pattern in bacterial DNA. They later learned to insert their own RNA guides, making the system programmable and turning it into a gene-editing tool now used in medicine.

What Did Claude Find?

The system Claude discovered is built around a reverse transcriptase, an enzyme that copies RNA into DNA. Bacteria use many such enzymes to fight viruses, according to Anthropic.

That specific enzyme had been recorded earlier in a jumbo phage, an unusually large bacteriophage. However, previous studies apparently missed the system's defining features. Claude seems to be the first to spot them.

One feature is a partner protein with an unknown role. The other is a long array of evenly spaced DNA repeats, similar in layout to a CRISPR array. CRISPR is mostly found in bacteria, while ART appears mainly in the phages that infect them.

Anthropic has named the three-part system array-associated reverse transcriptases, or ART. The Claude agent that found it recorded its surprise while reading the raw DNA.

Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.

We don’t yet understand what this system does, but only a handful of known…

— Anthropic (@AnthropicAI) September 23, 2026

How Did The Claude Agent Find It?

The agent was one of about 950 Claude agents performing the same search. Anthropic's scientists gave it a single prompt: to find new reverse transcriptases in a large DNA database.

Over 21 hours, the agents used 210 million tokens, the text units AI models process. They collected more than 200,000 reverse transcriptases and identified 3,500 new candidate systems.

The agents then narrowed that set to the 20 most promising ones and wrote reports for human review. According to Anthropic, such analysis can take an expert scientist weeks to months.

“While combing through the raw DNA sequence near the RT, the agent 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?!”

The agent that spotted ART worked much like a human scientist. It counted the repeats, measured their spacing, compared the layout to known systems, and checked the literature for earlier reports.

Anthropic's researchers provided only the prompt and performed the lab work themselves. The agents used their own judgment to decide which leads to follow.

That lab work took place at the company's Bay Area facility, which does not handle human-infecting pathogens.

What Has the Lab Found So Far?

Initial experiments at that lab provide an early hint about how ART might work. The team found that ART's repeat array is transcribed into a set of distinct short RNAs.

That resembles CRISPR, where each short RNA acts as a guide pointing the system to a target. Anthropic says the result suggests a similar process may occur with ART.

ART's combination of features is also rare. According to Anthropic, only a handful of other known systems share it.

All of those are programmable and can act on DNA, for example by cutting, copying, or pasting it. Besides CRISPR, several are being developed as tools.

However, Anthropic has not shown that ART can do any of this. The company says the system's primary function is still unknown and further experiments are under way.

What Does Anthropic’s CEO Make of It?

Anthropic CEO Dario Amodei said on X that the company suspects ART could be a new gene-editing mechanism. He added that its function, usefulness, and significance are not yet clear.

“It’s easy to dismiss this as a one-off or curiosity, but we’ve repeatedly seen a pattern where AI performance in new intellectual domains goes from weak to superhuman in a matter of a few years,” he said.

Amodei noted that a Stanford team recently found a reverse transcriptase system with a non-coding array. According to him, it is in some ways similar to ART, but the two evolved independently.

Cool that an AI agent found the new CRISPR like system by reading the raw genetic code like plain text. It was a lucky find because the agents missed it on ten consecutive runs when in bio software.

I guess our bodies needs to be in plain text and then we won't die. https://t.co/SUhxrZPg5J pic.twitter.com/ixNhaiYab5

— Bryan Johnson (@bryan_johnson) September 24, 2026

Is This the Next CRISPR?

Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, reviewed the preprint, an early paper that has not been peer-reviewed. He called the RNA-repeat arrays intriguing and worth further investigation.

Kevin Blake, a microbiologist at Washington University School of Medicine, was more skeptical. He told Al Jazeera that CRISPR in nature is very different from CRISPR as a technology.

He also noted that countless CRISPR-like sequences remain uncatalogued, because millions of bacterial species have yet to be studied.

“There’s nothing to indicate this is a rival to CRISPR-the-technology, or could be developed into any kind of therapeutic or practical application,” he commented.

Anthropic says further experiments are under way to determine how ART functions. The results will show whether its repeats point to a new programmable tool or a quirk of phage biology.