1,000 Claude Agents Ran 21 Hours to Find Anthropic's New Enzyme

Anthropic says its Claude model has "autonomously discovered" a new enzyme system with parallels to the gene-editing machinery behind Crispr, according to The Verge. The finding is the first result to come out of the company's newly launched wet lab, arriving as Anthropic prepares to go public and positions Claude as a tool for original scientific work rather than just text generation.

What happened over the 21-hour run?

Nearly 950 Claude agents worked in parallel for 21 hours, consuming 210 million tokens as they searched a large database of DNA sequences, Anthropic told The Verge. One agent flagged an unusual repeating pattern in the dataset and surfaced it for human review. Anthropic says its scientists' role was limited to writing the initial prompt and running the follow-up lab work — the pattern-spotting itself was left to the agents.

What did Claude actually find?

Further analysis and bench testing on the flagged sequence turned up what Anthropic describes as "a previously uncharacterized enzyme system found in bacteriophages," the viruses that infect bacteria, per The Verge's reporting. Anthropic says it is still working to determine what the enzyme system does. The company has not published a peer-reviewed paper laying out the enzyme's function or mechanism, based on the available reporting.

Why is Anthropic comparing this to Crispr?

Crispr, the gene-editing system adapted from a bacterial immune defense, transformed biology and medicine after its discovery. Anthropic's comparison rests on the fact that its new enzyme system was also found in bacteriophage-related sequences, the kind of source material that has previously yielded gene-editing tools. The Verge notes it "remains unclear" whether the newly flagged enzyme system will prove to have Crispr-like utility — the comparison is Anthropic's framing of the discovery's category, not yet an established fact about its function.

Anthropic said it wanted to disclose the result before that verification work is finished. "It felt it was 'important to share such findings early, both to demonstrate Claude's capabilities and to give the broader community insight into what we're working on,'" the company told The Verge.

Why does this matter now?

The timing matters because Anthropic is using the result to argue Claude can do more than summarize or write — that it can generate a genuine scientific lead with minimal human steering. That argument lands as the company readies an IPO and competes with rivals making similar claims about AI-driven science.

What to watch

• Whether Anthropic or outside researchers publish a peer-reviewed characterization of the enzyme system's function
• Whether independent labs can replicate or build on the bacteriophage sequence Claude flagged
• Whether Anthropic discloses more detail on how the 950-agent search was structured and validated
• How the claim factors into Anthropic's public offering narrative in the coming months

Anthropic's own account, as relayed by The Verge, is the only public record of the experiment's design and results so far; no independent lab has confirmed the enzyme system's properties.