Claude enzyme CRISPR: The Atlantic says the jury is still out on whether it is an enzyme

The Claude enzyme CRISPR case gained a new chapter on October 4. The Atlantic published a piece that puts Anthropic's claim in perspective: that Claude found, in under 24 hours, an "enzyme system" resembling CRISPR. The new angle is not the September announcement. It is public skepticism from scientists, including someone who collaborates with the company.

In AI's Real Gift to Science, Hana Kiros notes that Anthropic CEO Dario Amodei has said AI could cure most diseases within ten years. In late September the company presented the finding as a first step. The Atlantic records that Anthropic was open about not knowing exactly what it had found, or whether the result — the first released from the biology lab opened this spring — was truly significant. Academics and the CEO of Eli Lilly nonetheless argued that an incremental result had been dressed up as a larger discovery.

What Anthropic claims

In its official announcement, Anthropic says Claude agents identified a system it calls ART, short for array-associated reverse transcriptases, in phages that infect bacteria. Wired and TechCrunch reported that about 950 agents searched sequences in 21.5 hours. The Atlantic adds that the database held nearly 2 billion DNA entries. The company describes properties "reminiscent" of CRISPR and says it does not yet understand what the system does. The technical report has not been peer-reviewed.

What researchers contest

Fyodor Urnov, the scientist who coined the term genome editing and is working on a CRISPR project with Anthropic, told The Atlantic that "the jury is out" on whether what Claude found is even an enzyme. That is his statement, not a published experimental verdict.

The New York Times had already reported, on September 27, that Mario Rodríguez Mestre, a computational biologist at the University of Copenhagen, says his group has studied for four years the enzymes and associated molecules Anthropic calls ARTs. Those findings are not yet published. Mestre told the paper the main question is not whether ARTs were already known — in his view, they were — but whether the model reasoned its way there or was partly guided by his work, shared in interactions with Claude. That remains his allegation, not an established fact.

In response to the Times, Anthropic said it was not aware of any previously published work describing the ART system it found, and that Claude is not trained on user transcripts, nor does the molecular biology team have that access. The Atlantic notes the company did not respond to a comment request for the October 4 piece.

What the report still preserves

Kiros does not dismiss using agents to sort biological data. Her point is different: by selling the finding as a step toward curing disease, Anthropic obscured what Claude appears to have done in practice — scan a volume of sequences that would take a human group far longer. Kendell Clement of the University of Utah tells The Atlantic that scale is the difference from tools already used to find gene-editing systems. Stirling Churchman of Harvard says there will be "magic where there's data," and less use in fields with fewer public sequences.

What is not confirmed is whether ART is a functional enzyme, whether it cuts, copies, or pastes DNA in a new way, and whether the path to it depended on unpublished research. Bench experiments and independent review are what can change that stage.

Sources

Transparency: This content was created, edited, or reviewed with the aid of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources for the full context.

Por GeekikiBot