Has AI arrived as a scientific discoverer? Claude’s new CRISPR-like discovery uncovers a hidden enzyme system that scientists had missed in bacterial DNA

A strange pattern in bacterial DNA has led Anthropic researchers to a previously unknown biological system. The company says Claude found the pattern while searching vast genetic databases for unusual reverse transcriptases. The discovery, array-associated reverse transcriptases, or ARTs, is still at an early stage. Researchers do not yet know what the system does inside a living cell.

ART is not a new gene-editing technology yet. No one has shown that it can edit DNA, target a chosen sequence or perform the kind of programmable genetic changes associated with CRISPR.

What is special is how the clue was found. Claude was not given a known biological mechanism and asked to explain it. Its agents searched for patterns that might have been overlooked.

A familiar enzyme was sitting beside something scientists could not explain

Reverse transcriptases are not new to biology. They are enzymes that use RNA as a template to make DNA. They appear in many biological systems, including viruses and bacteria.

The particular reverse transcriptase involved in this work had already been identified in a jumbo bacteriophage. The surprise came from what surrounded it.

Claude found another gene next to the reverse transcriptase. Scientists could not assign that gene a known function. Beyond it sat a long array of repeated DNA sequences.

The arrangement stood out because the pieces appeared together. The array contained between three and 21 repeats. Each repeat was between 15 and 49 nucleotides long.

Any one of those features might not have been enough to attract attention. Together, they suggested that the reverse transcriptase, the unknown gene and the DNA array could be parts of the same system.

That was the lead researchers decided to investigate.

Claude searched billions of protein clusters before finding the pattern

The scale of the search is one of the more unusual parts of the discovery.

Anthropic says its researchers started with more than 1.9 billion protein clusters and asked Claude to search for interesting reverse transcriptases. The agents eventually narrowed the field to about 200,000 candidate reverse transcriptases.

They then looked more closely at roughly 3,500 partner gene families.

The work involved about 949 Claude agent sessions running for around 21 hours. In total, the system processed roughly 210 million tokens, according to Anthropic. The agents ultimately produced 19 reports that human researchers could examine.

Those numbers do not mean Claude independently discovered a biological system from scratch. Humans designed the research task, supplied the data and evaluated the resulting evidence. The laboratory work also remained essential.

What changed was the amount of searching that could happen before a scientist had to decide which clue deserved attention.

A researcher can spot an unusual sequence arrangement. An AI system can examine huge numbers of possible arrangements and look for combinations that would be difficult to inspect manually.

In this case, one of those comparisons exposed the repeated DNA next to the reverse transcriptase.

The CRISPR connection comes from the shape of the DNA array

The ART system drew more attention because its repeated DNA has an architectural resemblance to CRISPR.

CRISPR systems are known for arrays of repeated sequences separated by variable DNA segments. Parts of these arrays can be transcribed into RNA molecules that help guide associated molecular machinery toward particular genetic material.

ART is not known to work that way.

The similarity is about organization, not demonstrated function. ART contains a reverse transcriptase, an adjacent accessory gene and a repeated DNA array. Early laboratory experiments also found that the array can be transcribed into separate short RNA fragments.

That observation gives scientists a reason to investigate the system further. It does not establish that those RNAs guide the enzyme.

There is currently no evidence in the material provided showing that ART can make targeted genetic changes or operate as a programmable gene-editing system.

That is the gap between an intriguing biological discovery and a useful biotechnology.

Scientists still do not know what ART actually does

The biggest question is also the most basic one: what happens when ART is operating inside a cell?

Researchers have found a recurring genetic arrangement and evidence that its DNA array is transcribed. They still need to determine what the resulting RNAs do, how the reverse transcriptase interacts with them and what role the neighboring gene plays.

The system could turn out to have a function that is completely different from CRISPR.

That possibility is important because biological resemblance can be misleading. Two systems can share an unusual molecular architecture while performing very different jobs.

The reverse transcriptase itself may also provide clues. If it uses one of the short RNAs as a template, that could point toward a previously unknown form of information transfer between RNA and DNA. If the RNA instead interacts with another component, the mechanism could look very different.

Those experiments have not yet been settled.

For now, ART should be viewed as a biological mystery rather than a finished gene-editing technology.

Its significance lies partly in the discovery itself and partly in the search process behind it. Claude examined genetic relationships at a scale that would be difficult for a small research team to reproduce manually. It found a combination of features that had not been the original focus of the investigation.