Anthropic’s Claude Found an Enzyme System—Scientists Question What It Really Discovered

Anthropic says Claude helped uncover a potentially new gene-editing mechanism in bacteriophages. Outside researchers say the finding remains unproven—and one scientist questions whether the underlying system was already known to the model.
Anthropic’s Claude Found an Enzyme System—Scientists Question What It Really Discovered

Anthropic’s Claude Found an Enzyme System—Scientists Question What It Really Discovered
Anthropic opened its Bay Area wet lab this spring, pairing Claude with human scientists conducting lower-risk biology experiments rather than letting the model operate equipment itself. The company says that setup quickly produced its first major result: a previously unknown enzyme system in bacteriophage DNA with features resembling CRISPR, the bacterial immune system now widely used for gene editing.

On Wednesday, Anthropic said Claude had searched the scientific terrain through roughly 950 agents, consuming 210 million tokens over 21 hours of concentrated work. Chief executive Dario Amodei framed the result carefully but ambitiously, calling it a “molecular machine” that “could represent a new gene editing mechanism,” while acknowledging that its function, usefulness and significance remain unclear.

That caveat has become central to the response. Eric Kauderer-Abrams, Anthropic’s life-sciences chief, said biology and medicine are the company’s main route for delivering AI’s benefits. But the findings were released in a company technical report, not a peer-reviewed journal, and Harvard Medical School microbiologist Philip Kranzusch said the evidence does not yet justify a breakthrough label: “It’s a wrinkle on what we know in the field.”

The scrutiny sharpened over the weekend when computational biologist Mario Rodríguez Mestre said his group had studied the enzyme system—called ARTs by Anthropic—for four years. His team had also used Anthropic models in its work and shared findings with the company. For Mestre, the key question is not whether ARTs were already known—“They were”—but whether Claude independently reasoned to the result or benefited from information embedded in later model training.

Anthropic’s announcement may still mark a striking demonstration of AI-assisted biological search. But the next test belongs to outside laboratories: establish what the system does, whether it is genuinely novel, and how much of the path to it was Claude’s alone.

https://foxvector.com

Write a comment