Claude Finds a New Enzyme System Featuring CRISPR-like Repeats

Meta description: Anthropic Claude found a CRISPR-like enzyme system, ART, after 950 agents spent 21 hours and 210 million tokens mining DNA sequences in a massive database.

Tags: Anthropic Claude, CRISPR, reverse transcriptase, AI biology, genome mining

Featured image: abstract DNA sequences and enzyme structures in a computational biology laboratory style, no people, logos, or text

Anthropic Claude autonomously discovered array-associated reverse transcriptases (ART), a novel enzyme system with CRISPR-like DNA repeats, in early results published Sep 23, 2026. The finding comes from Anthropic’s new life sciences research group and laboratory, which explores DNA datasets, identifies uncharacterized protein families, generates hypotheses at scale, and tests them through experiments, with only high-level direction from scientists. It shows a path from prompt to laboratory evidence in which agents search biological anomalies rather than only summarize them.

Biological Tools Often Emerge From Odd DNA Patterns

Many breakthroughs began with odd molecular machines. Restriction enzymes, proteins that cut DNA at specific short sequences, were found in bacterial immune systems, where they destroy invading viral DNA; using them to cut and splice genes launched the biotechnology industry. Taq polymerase, an enzyme that copies DNA at high temperatures, was identified in a bacterium in a Yellowstone hot spring and became the basis for PCR, used in much modern diagnostics. CRISPR began as an unusual repeat sequence in certain bacteria and now underlies gene editing-based medicines.

In recent years, researchers discovered many more reverse transcriptases (RTs), enzymes that copy RNA into DNA, most in bacteria, where they act as part of the immune system. Nearly all RT families came from genome mining: searching sequence databases for uncharacterized genes, noticing unusual ones, and working out their function. For an expert scientist, that analysis can take weeks to months.

Spring 2026: Anthropic Builds a Claude Biology Lab

In spring 2026, Anthropic formed a research group to see whether general AI models can systematize and accelerate discoveries. It built a Bay Area lab, a typical molecular biology lab, and one team covering training Claude in biology through lab experiments, with agents collaborating with humans at every step. The lab conducts only BSL-1 and BSL-2 work and does not handle pathogens that can infect humans; all lab work is performed by human scientists. Anthropic has experimented with AI lab acceleration through the Model Hardware Standard but says that approach is less conducive to ad hoc molecular biology workflows.

The team studies unusual proteins and uses computational approaches to read DNA, interpret evolution, and select systems for characterization. Before Anthropic, its work helped understand CRISPR evolution and regulation, discover enzymes for next-generation cell and gene therapies, and identify DNA anomalies such as human pathogenic variants. It sits in Anthropic’s life sciences organization alongside drug discovery and training Claude in biology and chemistry.

Typical workflows have Claude survey protein families, read literature, reproduce public-data results to check methods, find unassigned family members or genomic neighbors, and write human-readable reports proposing functions and evidence. Claude critically evaluates the evidence; most candidates are eliminated, and a survey may end with one candidate worth testing, or none. Survivors are tested in standard laboratory strains and characterized biochemically and structurally, with Claude helping interpret data. The team uses Claude Science and Claude Code, the same tools available to any scientist, and sometimes a custom harness coordinating many parallel sessions. Hundreds to thousands of candidate reports per campaign teach Anthropic what is worth testing, and those lessons are fed back to shape Claude’s scientific taste.

September 23, 2026: 950 Agents Find ART

Anthropic gave Claude a prompt to search a massive DNA database for RTs; humans provided only the initial prompt and lab work. Claude agents gathered over 200,000 RTs, picked out 3,500 new candidate systems, and narrowed those to the 20 most compelling candidates analyzed in human-readable reports. Claude noticed an unusual RT family and examined it in greater detail. After 21 hours spent searching the data by roughly 950 agents using 210 million tokens, one agent spotted a repeating DNA pattern next to an odd-looking RT. It 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 counted repeats, measured spacing, compared known RT systems, searched the literature for any previous report of the pattern, and filed a report for human review. Lab testing identified ART in bacteriophages, the viruses that infect bacteria. The underlying RT, found in a jumbo phage, had been known, but Claude appears to be the first to notice ART’s defining features: an associated array of non-coding DNA sequences and an additional accessory protein of unknown function.

ART is found mainly in bacteriophages and consists of three parts: the RT, a partner gene beside it, and a long array of evenly spaced DNA repeat sequences. The layout resembles a CRISPR array, which holds a bank of different RNA sequences that make CRISPR-Cas systems programmable biotechnological tools. Anthropic’s first experiments show that the ART array is also expressed as distinct short RNAs, suggesting something analogous may be at play. ART has characteristics only ever found together in a handful of other systems, all of which are programmable and perform operations like cutting, copying, and pasting DNA. Beyond CRISPR, which has already transformed science and medicine, several other such systems are now in development as promising tools.

Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute, reviewed the pre-print: “This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research.”

Anthropic says it shared the finding early both to demonstrate Claude’s capabilities in AI-driven hypothesis generation and to give the broader community insight. ART’s primary function remains unknown, and experiments are underway to determine how it works. What to watch next is whether ART can be understood well enough to become a programmable DNA tool, and whether Anthropic’s Life Sciences Verification Program, which gives life science professionals access to Claude Mythos, Opus, and Sonnet models with refined safeguards more permissive for biology-related work, expands this workflow. Anthropic Science features AI-assisted discoveries, practical workflows, and field notes across the sciences, and the company wants to work with other scientists to extend the approach to genomics and other fields.