Biotech

Anthropic Says Claude Found a CRISPR-Like System in 21 Hours

co-produced by AI24. September 2026, 15:41

Illustration einer DNA-Doppelhelix aus Blumen und Blättern als Sinnbild für KI in der Genforschung — © Nidia Dias / Google DeepMind / Pexels

A.I. is now doing basic research, at least if you take Anthropic’s word for it. The company behind Claude has just announced that its A.I. model discovered a previously unknown enzyme system in the DNA of bacteriophages, one that resembles the famous gene-editing tool CRISPR. About 950 A.I. agents needed roughly 21 hours to find it, Anthropic wrote in a blog post. What the system actually does, however, nobody knows yet.

What Claude Found

Anthropic calls the discovery “array-associated reverse transcriptases,” or ART. The system has three parts: an enzyme of the reverse transcriptase type, a partner gene with an unknown function and a long array of repeating DNA sequences. Those repeats are what resemble CRISPR, the natural bacterial defense system that gave rise to today’s most important gene-editing tool.

In the lab, Anthropic’s researchers confirmed that the DNA array is expressed as short, distinct pieces of RNA. That pattern is familiar from programmable systems that can make targeted changes to DNA. Whether ART actually does that remains open. “We don’t yet understand what this system does,” Anthropic concedes. The company has released a preprint and a technical report and is looking to collaborate with outside researchers.

How the Search Worked

Humans wrote the initial search prompt, and Claude agents handled the rest. According to The Next Web, they screened about 1.9 billion protein clusters and came across roughly 200,000 reverse transcriptases. From those, they filtered 3,500 candidates, then 20 especially promising ones, and wrote readable reports on each. One agent eventually noticed the distinctive repeat pattern sitting right next to one of the enzyme genes. In total, the search used about 210 million tokens.

All lab work was done by humans. Anthropic has run its own molecular biology lab in the Bay Area since the spring, working only at the low biosafety levels BSL-1 and BSL-2 and handling no pathogens that infect humans. “Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today,” CEO Dario Amodei said, according to TechCrunch. The discovery was made “mostly, though not entirely, by Claude,” he said.

Excitement and Skepticism

A prominent name is lending support: Feng Zhang, one of the pioneers of CRISPR technology at MIT and the Broad Institute, called the discovery “genuinely intriguing” and worth further investigation. Stanley Qi, a bioengineer at Stanford University, told Al Jazeera it was an “incredibly exciting” example of how A.I. can detect complex biological patterns.

Others are more cautious. Kevin Blake, a microbiologist at Washington University, said there was “nothing to indicate this is a rival to CRISPR-the-technology.” Countless CRISPR-like sequences in millions of unstudied bacterial species remain undiscovered anyway, he noted. There are further caveats:

Not entirely new: The enzyme itself was already known. Claude’s contribution was spotting the overlooked link to the DNA array and the partner protein. According to Amodei, Stanford researchers had already found a system “in some ways similar” to the one Claude found.
Hard to reproduce: According to The Next Web, Anthropic reran the search ten times, and not a single run found the pattern again. The discovery depended partly on chance in how the agents behaved.
Function unclear: Further experiments will have to show whether ART can cut or edit DNA.

The Race for A.I. Science

Anthropic isn’t the only A.I. lab pushing into biology. Google DeepMind transformed protein structure prediction with AlphaFold, earning Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry. OpenAI and numerous start-ups are also working on A.I. for drug discovery. In the spring, Anthropic acquired the biotech start-up Coefficient Bio for about $400 million to build up its life sciences expertise.

The ambitions are huge: Amodei has repeatedly predicted that A.I. could “cure most diseases in 5-10 years.” The ART discovery is still far from that. But it shows that A.I. agents can find connections in vast amounts of data that humans have missed.

What Is CRISPR, and Why Does It Matter?

CRISPR stands for “clustered regularly interspaced short palindromic repeats,” meaning short, repeating DNA sequences arranged at regular intervals. Originally, CRISPR is an immune system in bacteria: when a virus (a so-called bacteriophage) attacks a bacterium, the bacterium stores a piece of the viral DNA between these repeats, like a mugshot archive. If the virus returns, the bacterium recognizes it from that copy, and an enzyme called Cas (such as Cas9) cuts the viral DNA at the right spot.

In 2012, French microbiologist Emmanuelle Charpentier, who began parts of her foundational research at the Max Perutz Labs in Vienna, and American biochemist Jennifer Doudna realized the system could be repurposed: with a custom guide RNA, Cas9 can be sent to almost any location in a genome and cut there. That makes it possible to switch off, repair or replace genes in plants, animals and humans. The two won the Nobel Prize in Chemistry for it in 2020.

Why it matters so much:

Precise and cheap: Genetic engineering used to be laborious, expensive and imprecise. CRISPR makes targeted genome edits possible in almost any lab, often within weeks.
Medicine: Casgevy is already an approved CRISPR therapy for sickle cell disease and beta thalassemia, including in the EU. More therapies for genetic diseases, cancer and infections are in development.
Agriculture and research: CRISPR helps breed more resilient crops and has become a standard tool in basic biological research.
Ethical limits: The technology also raises big questions, most visibly since a Chinese researcher edited the genomes of embryos in 2018 and sparked global outrage.

That is why every report of a new CRISPR-like system electrifies the field: each new natural tool could one day enable more precise, smaller or more versatile gene editors. Whether ART is one of them remains to be seen.