Anthropic reports that Claude has uncovered a previously unidentified enzyme system embedded within the DNA of viruses that infect bacteria. While past research had documented the central enzyme, it appears to have overlooked the surrounding genetic structures.
Anthropic released these findings on September 23, presenting them as key evidence that artificial intelligence can now identify patterns that human researchers miss. But what does this discovery actually signify?
Why Should Anyone Outside a Lab Care?
Numerous medical breakthroughs started when a scientist observed an unusual biological anomaly. For instance, researchers’ utilization of restriction enzymes found in bacteria helped pave the way for the modern biotechnology industry.
Similarly, an enzyme extracted from a Yellowstone hot spring provided the foundation for PCR, the DNA-replication method that powers much of contemporary medical diagnostics.
Claude has now taken that initial observation step, guided only by broad instructions from Anthropic’s scientific team. Anthropic CEO Dario Amodei suggests this points to a broader trend. Back in 2023, AI models routinely struggled with high-school-level math.
By late 2026, he notes, these systems are beginning to crack some of the most complex open problems in the scientific field. He contends that biology-focused AI is advancing along a similar trajectory.
Should his prediction hold true, Amodei asserts that accelerated discovery could unearth novel drug targets and innovative treatment classes. While this would not shorten clinical trials, it could feed a significantly higher volume of promising candidates into the drug development pipeline.
He has previously expressed that AI could assist in eradicating the majority of diseases within five to ten years, though he describes that objective as only narrowly feasible.
At present, human researchers still conduct every physical experiment, and the precise biological function of the newly found system remains a mystery.
What Exactly Is CRISPR?
Bacteriophages, commonly known as phages, are viruses that continuously target bacteria. In response, many bacteria defend themselves using a mechanism known as CRISPR.
CRISPR functions essentially as a biological memory archive. Bacteria store short fragments of DNA from previous viral intruders between repeating sequences—a region referred to by scientists as an array.
Each DNA snippet is transcribed into a short strand of RNA. This RNA then directs a specialized protein to locate and cleave matching viral DNA if the identical pathogen strikes again.
Initially, scientists identified CRISPR simply as an unusual repeating pattern within bacterial DNA. They later discovered how to substitute custom-designed RNA guides into the system, transforming it into a programmable gene-editing instrument widely utilized in medicine today.
What Did Claude Find?
The system discovered by Claude centers around a distinct type of enzyme called a reverse transcriptase, which transcribes RNA back into DNA. According to Anthropic, bacteria deploy numerous such enzymes to ward off viral infections.
While researchers had previously logged this particular enzyme within a jumbo phage—an exceptionally large virus targeting bacteria—earlier studies apparently missed the system’s defining characteristics. Claude appears to be the first to flag them.
Among these features is a partner protein whose function remains entirely unknown. The other is an extensive array of evenly spaced DNA repeats, configured in a manner strikingly similar to a CRISPR array. Whereas CRISPR is predominantly found in bacteria, ART primarily appears in the phages that infect them.
Anthropic has designated this three-component configuration as array-associated reverse transcriptases, or ART. The specific Claude agent that made the discovery logged its astonishment directly as it scanned the raw genetic data.
How Did The Claude Agent Find It?
This agent operated alongside roughly 950 other Claude instances performing the same search. Anthropic researchers initiated the process with a single prompt, instructing Claude to locate novel reverse transcriptases within a massive DNA database.
Over the course of 21 hours, the agents consumed 210 million tokens—the fundamental text blocks processed by AI models. They compiled over 200,000 reverse transcriptases and isolated 3,500 new candidate systems.
The agents subsequently filtered that pool down to the 20 most compelling candidates and drafted comprehensive reports for human evaluation. Anthropic notes that this caliber of analysis typically demands weeks or months of effort from an expert scientist.
“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 responsible for finding ART operated much like a human researcher would: it tallied the repeats, measured the spacing between them, cross-referenced the layout against established systems, and searched academic literature for prior documentation.
Anthropic’s team provided solely the initial prompt and executed the subsequent laboratory work independently. The AI agents relied on their own analytical judgment to determine which leads warranted pursuit.
That laboratory work was conducted at the company’s Bay Area facility, which does not handle pathogens capable of infecting humans.
What Has the Lab Found So Far?
Initial experiments conducted at the facility offer preliminary insights into how ART might function. The research team observed that ART’s repeat array is transcribed into a collection of distinct, short RNA molecules.
This mirrors the behavior of CRISPR, where individual short RNAs act as guide molecules directing the system toward a specific target. Anthropic states that this outcome hints at a comparable process occurring within ART.
Furthermore, the specific combination of features present in ART is exceptionally rare. Anthropic reports that only a small number of known systems share this profile.
All of those comparable systems are programmable and capable of acting upon DNA—such as by cutting, copying, or pasting genetic sequences. Aside from CRISPR, several of them are currently being cultivated as functional tools.
Nevertheless, Anthropic has not yet demonstrated that ART possesses these capabilities. The company emphasizes that the primary function of the system remains unknown, and additional experiments are currently underway.
What Does Anthropic’s CEO Make of It?
Dario Amodei, Anthropic’s chief executive, stated on X that the organization suspects ART could represent a novel gene-editing mechanism. In the same post, he cautioned that its practical function, utility, and broader significance remain undetermined.
“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 pointed out that a research team at Stanford recently identified a reverse transcriptase system featuring a non-coding array. According to him, that system shares certain similarities with ART, though the two evolved completely independently of one another.
Is This the Next CRISPR?
Feng Zhang, a prominent CRISPR pioneer at MIT and the Broad Institute, reviewed the preliminary research paper—known as a preprint, which has not yet undergone peer review. He characterized the RNA-repeat arrays as fascinating and worthy of deeper exploration.
Conversely, Kevin Blake, a microbiologist at the Washington University School of Medicine, expressed greater skepticism. He told Al Jazeera that naturally occurring CRISPR systems differ significantly from CRISPR as an engineered technology.
He also highlighted the fact that countless CRISPR-like sequences remain uncatalogued simply because millions of bacterial species have yet to be thoroughly 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 confirms that further experiments are actively underway to determine the precise operational mechanics of ART. The resulting data should clarify whether its repeat sequences point toward a novel programmable tool or simply represent a unique quirk of phage biology.
Subscribe to our YouTube channel to watch leaders and journalists provide expert insights
Frequently Asked Questions
What did Claude discover?
Claude discovered a previously unknown enzyme system in bacteriophage DNA, named array-associated reverse transcriptase (ART). It consists of a reverse transcriptase enzyme, an unknown partner protein, and a long array of evenly spaced DNA repeats resembling a CRISPR array.
How was the discovery made?
Anthropic deployed roughly 950 Claude agents to search a massive DNA database for new reverse transcriptases. Over 21 hours, the AI processed 210 million tokens, evaluated over 200,000 reverse transcriptases, and flagged 3,500 new candidate systems before narrowing them down for human review.
Do scientists know what ART does?
No, the exact function of the ART system remains unknown. While preliminary lab work shows its repeat array is transcribed into short RNAs—similar to CRISPR—further experiments are required to determine its primary purpose or practical application.
Is ART a new gene-editing tool like CRISPR?
It is too early to tell. While experts like Feng Zhang find the arrays intriguing, microbiologist Kevin Blake and other skeptics note there is currently no evidence that ART can serve as a rival to CRISPR-the-technology or be developed into a therapeutic tool.
