Anthropic says hundreds of Claude AI agents identified an uncharacterized biological system while searching vast DNA databases. Its function remains unknown, but the finding raises a larger question: could AI help scientists uncover biological mechanisms humans have overlooked?
THE UNIVERSAL RECORD
Sourced reporting. No opinions.
Brad Socha | September 30, 2026 | 9:59 AM EST
Artificial intelligence may be crossing an important boundary in science: from organizing what humans already know to helping researchers discover biological systems they have not previously recognized.
On September 23, Anthropic reported that approximately 950 Claude AI agents spent about 21 hours searching enormous collections of DNA sequences and identified an unusual enzyme system with features resembling CRISPR, the naturally occurring biological machinery that helped transform modern gene editing.
The company named the newly characterized family array-associated reverse transcriptases, or ART. According to Anthropic, its agents examined more than 200,000 reverse transcriptases, identified approximately 3,500 candidate systems and narrowed their investigation to 20 for detailed analysis.
One agent noticed an unusual arrangement of genetic material in bacteriophages, viruses that infect bacteria. Beside a reverse transcriptase gene was a long array of regularly spaced DNA repeats accompanied by another gene. The arrangement resembled certain structural features of CRISPR systems, prompting further investigation.
The finding is scientifically intriguing, but an important distinction remains: researchers have not established that ART functions like CRISPR or can be used to edit genes. Its natural biological purpose is still unknown.
The larger story concerns what AI might eventually discover within the enormous quantities of biological information scientists have already collected.
Why CRISPR Changed Biology
CRISPR stands for clustered regularly interspaced short palindromic repeats. Although the name is complicated, its natural purpose is relatively straightforward.
Bacteria and archaea evolved CRISPR-associated defence systems that can help protect them against invading genetic material, including viruses. Some microorganisms retain fragments of viral DNA, effectively creating a molecular record that can help them recognize related threats during future encounters.
Scientists discovered that components of certain CRISPR systems could be adapted to target specific DNA sequences.
In 2012, researchers Emmanuelle Charpentier and Jennifer Doudna demonstrated how the CRISPR-Cas9 system could be programmed to cut DNA at predetermined locations. Their work contributed to a transformation in genetic engineering and earned them the 2020 Nobel Prize in Chemistry.
The implications extend far beyond microorganisms.
CRISPR-based technologies allow scientists to alter genetic instructions within living cells. Depending on the method, researchers can disable particular genes, change selected DNA sequences or investigate how genes contribute to disease.
The technology has already reached human medicine. Approved CRISPR-based treatments for sickle cell disease and transfusion-dependent beta thalassemia involve collecting a patient’s blood-forming stem cells, editing them outside the body and returning them following preparatory treatment.
Scientists are also investigating gene-editing approaches for inherited disorders, cancer research, drug development and agriculture. These applications vary considerably in maturity, and many remain experimental.
This history explains why unfamiliar biological machinery attracts attention.
CRISPR began as a naturally occurring system in microorganisms. Only after years of research did scientists understand enough about its operation to adapt selected components into programmable tools.
Discovering another previously uncharacterized system could eventually expand that scientific toolkit, but only if experiments establish what the system actually does.
What Claude Found, and What Remains Unknown
Anthropic’s AI agents were instructed to search genomic databases for interesting examples of reverse transcriptases, enzymes that copy genetic information from RNA into DNA.
While examining the genetic material surrounding one such enzyme, Claude identified three components appearing together: the reverse transcriptase, a neighbouring partner gene and an array of evenly spaced DNA repeats.
In simpler terms, the AI noticed several pieces of biological machinery arranged in a pattern unusual enough to suggest they might work together as a system.
Anthropic’s laboratory subsequently confirmed that the repeat array produces distinct short RNA molecules. That finding supports the computational observation that the genetic arrangement is biologically active, but it does not establish its function.
There are also important limitations to the discovery.
The underlying reverse transcriptase had appeared in earlier research. Anthropic’s reported contribution was recognizing its association with the surrounding genetic components. Some independent specialists have also questioned whether ART should be described as a CRISPR system at all, noting that structural resemblance does not establish shared biological function.
A separate question concerns how independently the discovery was made.
Computational biologist Mario Rodríguez Mestre of the University of Copenhagen has said his research group was already investigating the same biological system and had shared unpublished information while using Claude. Anthropic has maintained that the model responsible for its finding was not trained on user transcripts.
The available public evidence does not establish that Claude obtained its finding from Mestre’s unpublished work. The dispute nevertheless illustrates why documenting research provenance will become increasingly important as AI participates in scientific discovery.
Anthropic’s technical findings remain preliminary and have not completed peer review.
Researchers must still determine what ART does naturally, whether its RNA molecules interact with the enzyme, what role the accompanying protein performs and whether independent laboratories can reproduce the results.
Only then could scientists assess whether the system might eventually offer useful applications.
That could include new ways to study genetic mechanisms or manipulate biological molecules, potentially contributing to biotechnology, medicine or agriculture. There is currently no evidence that ART can edit human DNA, treat disease or perform the functions of existing CRISPR-based technologies.
The distinction matters because the larger development is not simply the discovery of one unusual enzyme arrangement.
Scientists have accumulated vast genomic and protein datasets containing biological relationships that remain poorly understood. AI agents can search these collections, compare patterns across organisms, examine existing research and identify candidates that might otherwise receive little attention.
Rather than only answering questions about established scientific knowledge, AI can increasingly help researchers decide which new questions are worth investigating.
That capability could eventually assist in identifying unfamiliar proteins, enzymes, molecular interactions or biological processes. Whether any particular finding becomes useful will depend on experimental evidence, reproducibility and a detailed understanding of its mechanism.
ART offers an early example of both the possibilities and limitations of that approach.
Claude identified an unusual genetic arrangement and helped direct researchers toward laboratory experiments. What the arrangement does, whether it represents a genuinely distinct biological mechanism and whether it has practical applications remain open questions.
The computer can point scientists toward something unusual. Biology still gets the final vote.
Sources:
Anthropic — Claude Discovers a Novel Enzyme System With CRISPR-Like Repeats
https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
Anthropic — Autonomous AI Agents Discover Reverse Transcriptases With Tandem Repeat Arrays
https://www-cdn.anthropic.com/9b93e3e956b98b8cb1525e036848164b9bd6f4f6.pdf
Nature — Anthropic’s AI Biolab Finds ‘CRISPR-Like’ DNA in Viruses. What’s Next?
https://www.nature.com/articles/d41586-026-03039-6
WIRED — Anthropic Says It Discovered a Crispr-Like System. Now What?
https://www.wired.com/story/anthropic-says-it-discovered-a-crispr-like-system-now-what/
Reuters — Anthropic Says Claude AI Helped Discover Novel Enzyme System
https://www.reuters.com/business/healthcare-pharmaceuticals/anthropic-says-claude-ai-helped-discover-novel-enzyme-system-2026-09-23/
National Human Genome Research Institute — CRISPR
https://www.genome.gov/genetics-glossary/CRISPR
Nobel Prize — The Nobel Prize in Chemistry 2020: CRISPR-Cas9
https://www.nobelprize.org/prizes/chemistry/2020/press-release/
U.S. Food and Drug Administration — FDA Approves First Gene Therapies to Treat Patients with Sickle Cell Disease
https://www.fda.gov/news-events/press-announcements/fda-approves-first-gene-therapies-treat-patients-sickle-cell-disease
The Irish Times/New York Times — Did Anthropic’s Artificial Intelligence Really Make a Scientific Discovery on Its Own?
https://www.irishtimes.com/world/2026/09/28/did-anthropics-artificial-intelligence-really-make-a-scientific-discovery-on-its-own/
About the Author
Brad Socha is the founder of The Universal Record, focused on sourced, factual global reporting. Coverage includes international news, geopolitics, technology, and major developments.