Anthropic reports that Claude has uncovered a previously unidentified enzyme system within the DNA of viruses that infect bacteria. While earlier research had documented the enzyme at its core, past studies appear to have overlooked its surrounding structures.
Anthropic released these findings on September 23, providing notable proof that artificial intelligence can now identify patterns that human specialists have missed. Even so, the exact implications of this discovery remain to be seen.
Why Should Anyone Outside a Lab Care?
A multitude of breakthroughs that transformed medicine started when a scientist observed an unusual phenomenon in nature. For instance, researchers’ utilization of restriction enzymes found in bacteria helped establish the biotechnology sector.
Additionally, an enzyme sourced from a Yellowstone hot spring served as the foundation for PCR, the DNA-replication technique that underpins much of contemporary medical diagnostics.
Claude has now accomplished a similar observation step guided only by broad instructions from Anthropic’s researchers. Anthropic CEO Dario Amodei contends this aligns with a broader trend. Back in 2023, AI models struggled with high school-level mathematics.
By late 2026, he notes, these systems have begun solving some of the most difficult open problems within the discipline. He maintains that biology-focused AI is tracking along a comparable trajectory.
Should his assessment prove correct, Amodei suggests accelerated discovery could expose novel drug targets and innovative treatment modalities. While this would not shorten clinical trial timelines, it could introduce a much higher volume of promising candidates into the drug development pipeline.
He has previously expressed that AI could assist in eradicating most diseases within a span of 5 to 10 years, though he characterizes that objective as just barely attainable.
At present, human researchers continue to execute every physical experiment, and the precise function of the new system is still undetermined.
What Exactly Is CRISPR?
Bacteria encounter continuous threats from viruses known as bacteriophages, or phages. To counter them, many bacteria employ a defense mechanism designated as CRISPR.
CRISPR functions akin to a biological memory repository. Bacteria preserve brief segments of DNA sourced from prior invaders between sequences of repeating units, a stretch referred to by scientists as an array.
Every individual snippet is transcribed into a short ribonucleic acid (RNA) molecule. That RNA subsequently directs a protein to locate and cleave matching viral DNA if the identical virus reappears.
Researchers initially identified CRISPR via an unusual repeating pattern inside bacterial DNA. They subsequently discovered how to substitute custom-designed RNA guides of their own making, converting the mechanism into a programmable gene-editing instrument frequently utilized in modern medicine.
What Did Claude Find?
The architecture uncovered by Claude centers around a distinct category of enzyme called a reverse transcriptase, which translates RNA back into DNA. According to Anthropic, bacteria deploy numerous such enzymes to ward off viral infections.
Investigators had previously cataloged this specific enzyme inside a jumbo phage—an exceptionally large virus that attacks bacteria. Nonetheless, past examinations apparently overlooked the defining characteristics of the system, making Claude seemingly the first to detect them.
One feature is an associated partner protein whose purpose is currently undefined. The other is an extended array of uniformly spaced DNA repeats arranged in a manner strongly resembling a CRISPR array. Whereas CRISPR predominantly resides within bacteria, ART appears primarily inside the phages that infect them.
Anthropic designates this three-component configuration as array-associated reverse transcriptases, or ART. The specific Claude agent that located it recorded its astonishment while parsing the raw genetic code.
How Did The Claude Agent Find It?
The agent operated as one of roughly 950 separate Claude instances pursuing the identical search objective. Anthropic researchers initiated the process with a single prompt instructing Claude to track down novel reverse transcriptases across an expansive DNA repository.
Across a 21-hour span, the agents consumed 210 million tokens—the microscopic segments of text analyzed by artificial intelligence models. They accumulated upwards of 200,000 reverse transcriptases and isolated 3,500 prospective candidate systems.
Subsequently, the agents winnowed that roster down to the 20 most compelling candidates and compiled summaries for human evaluation. Anthropic points out that an equivalent analysis typically requires weeks to 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 identifying ART functioned in a manner closely mirroring human scientific inquiry. It tallied the repeats, gauged their spacing intervals, contrasted the structural layout against established systems, and scanned existing literature for prior documentation.
Anthropic’s researchers provided solely the initial prompt and executed the wet-lab procedures independently. The agents relied on their internal discretion to determine which potential leads merited further pursuit.
The physical laboratory work occurred at the corporation’s Bay Area facility, which does not handle pathogens capable of infecting humans.
What Has the Lab Found So Far?
Initial experiments conducted at that facility provide a preliminary indication regarding ART’s potential operational mechanics. The research group established that the repeat array belonging to ART is transcribed into a collection of distinct, short RNA molecules.
This characteristic mirrors CRISPR, wherein individual short RNAs serve as guides directing the system toward a designated target. Anthropic notes this outcome implies a comparable dynamic may be occurring within ART.
Furthermore, the specific blend of features exhibited by ART is uncommon. Data from Anthropic indicates that only a restricted number of other recognized systems possess these traits.
Every single one of those comparable systems is programmable and capable of acting upon DNA—such as by cleaving, replicating, or splicing it. Alongside CRISPR, multiple variants are actively being optimized as utility instruments.
Even so, Anthropic has not demonstrated that ART is capable of executing these functions. The organization emphasizes that the primary purpose of the system remains unknown, and additional experimentation is currently underway.
What Does Anthropic’s CEO Make of It?
Amodei, the chief executive of Anthropic, posted on X that the enterprise suspects ART could represent a novel gene-editing mechanism. Within the same message, he observed that its functional capacity, practical utility, and ultimate significance are not yet definitive.
“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 team at Stanford recently identified a reverse transcriptase system featuring a non-coding array. According to his assessment, that system bears certain similarities to ART, though the two evolved independently of one another.
Is This the Next CRISPR?
Feng Zhang, a prominent CRISPR pioneer situated at MIT and the Broad Institute, examined the preprint—an un-peer-reviewed preliminary research paper. He characterized the RNA-repeat arrays as fascinating and worthy of deeper exploration.
Conversely, Kevin Blake, a microbiologist affiliated with the Washington University School of Medicine, expressed greater skepticism. He conveyed to Al Jazeera that natural CRISPR environments differ substantially from CRISPR implemented as a technology.
He additionally remarked that countless CRISPR-like sequences persist uncatalogued due to the millions of bacterial species that have yet to be investigated.
“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 states that follow-up tests are ongoing to decipher how ART operates. The resulting data should clarify whether its repeat structures point toward a newly programmable instrument or simply represent an anomaly of phage biology.
Subscribe to our YouTube channel to watch leaders and journalists provide expert insights