Claude Just Discovered a New Enzyme System Using 950 Agents!

Hey everyone, it's Shii-chan! I found some genuinely surprising news today: AI discovered a brand-new enzyme system! Let's get into it.

Anthropic News anthropic.com

What was announced?

This news comes from Anthropic's News. Anthropic's life sciences research team, founded this past spring, used Claude to autonomously discover a novel enzyme system called ART, short for array-associated reverse transcriptases.

ART is made of three parts: a reverse transcriptase, which copies RNA into DNA; a partner gene whose function isn't yet known; and evenly spaced, non-coding DNA repeats. The arrangement of those repeats closely resembles CRISPR arrays, and early experiments suggest they're expressed as distinct short RNAs too, just like CRISPR. The reverse transcriptase itself was already known, but nobody had previously noticed the defining combination of nearby DNA repeats and an unknown partner protein.

Why it matters

Restriction enzymes, Taq polymerase, and CRISPR — discoveries of nature's odd molecular machinery have repeatedly reshaped biomedicine. Showing that more "programmable" enzyme systems like ART might still be out there points toward new tools capable of cutting, copying, and pasting DNA.

Anthropic's life sciences research team was set up to test whether AI models can systematize and accelerate biological discovery itself. The team includes scientists who train Claude on biology, researchers who run the lab experiments, and specialists in protein evolution analysis, with backgrounds spanning research on CRISPR system evolution and regulation, discovering enzymes for next-generation cell and gene therapies, and building tools to spot unusual DNA patterns.

What changes

What's striking about this discovery is the sheer scale. About 950 Claude agents ran in parallel over 21 hours, using 210 million tokens to collect over 200,000 reverse transcriptases, narrow them down to 3,500 new candidate systems, and finally settle on the 20 most compelling candidates. Doing a search at that scale with human researchers alone would take an enormous amount of time.

The research process runs through four steps:

• Claude reads relevant literature and reproduces already-established results from public data to confirm its methods are sound.
• It searches for gene families or genomic neighborhoods that don't fit known systems, and writes a human-readable report with its functional hypothesis and supporting evidence.
• Experts critically evaluate the hypothesis, and most candidates get filtered out at this stage.
• Only the surviving candidates move on to lab validation.

For ART, Claude noticed an unusual reverse transcriptase lineage, visually spotted the DNA repeat pattern, searched the literature while comparing it to known CRISPR systems, and reported discovering a new biological system. In the moment of discovery, the Claude agent left this note:

[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array … that's a CRISPR-like … repeat array?!

Dive Deep

In the lab validation stage, the protein gets expressed in standard laboratory strains and characterized biochemically and structurally. Claude helps interpret the results, and what's learned feeds back into how future searches are directed.

Anthropic's lab in the Bay Area only handles BSL-1 and BSL-2 low-risk work and doesn't work with pathogens that can infect humans, and all lab work is carried out by human scientists — the team judged that AI automation isn't yet a good fit for today's molecular biology workflows.

Feng Zhang, a CRISPR gene-editing pioneer and professor at MIT and the Broad Institute, commented on the discovery:

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.

He also said he hopes this work encourages more scientists to explore how AI can support their own research.

Wrap-up

• Claude autonomously discovered ART, a novel enzyme system with CRISPR-like repeats
• About 950 agents spent 21 hours and 210 million tokens narrowing 200,000+ reverse transcriptases down to 3,500 candidates, then 20 finalists
• The research process runs through reading literature, generating hypotheses, expert review, and lab validation
• All lab validation is done by human scientists in a BSL-1/BSL-2 low-risk facility
• CRISPR pioneer Feng Zhang praised the potential of AI agents contributing to biological discovery

If you're curious how far AI can go in contributing to scientific discovery, or interested in where life sciences and AI intersect, this one's worth a read!