Scientists praise Anthropic's fast Claude-led enzyme search but want lab proof of a Crispr-like system

WIRED reports on Anthropic's September 23 announcement that Claude identified an enzyme system "reminiscent of Crispr." About 950 agents searched genomic databases for novel reverse transcriptases, enzymes that copy RNA into DNA, and narrowed the search to an unusual family associated with repeated DNA arrays in jumbo phages. Anthropic calls these ART, for array-associated reverse transcriptases. The search reportedly took about 21.5 hours.

The evidence so far includes one physical experiment described in a technical report that has not been peer reviewed. Anthropic says it does not yet understand what the system does. The original announcement is covered in Anthropic says its biology lab has already found something big.

Experts quoted by WIRED were divided. UC Berkeley's Fyodor Urnov complimented Anthropic's disclosure. Stanford's Le Cong noted the publicity came ahead of experiments still in the queue. Gladstone Institutes' Seth Shipman credited the speed of the search rather than any autonomy in the discovery. Texas A&M's Jason Gill had identified the reverse transcriptase itself in 2021, so the novelty may lie in the repeat arrays that suggest a Crispr-like organization. Sources also distinguished the system from retrons, a different class of bacterial elements.

Anthropic told The New York Times that Claude was not trained on user transcripts, responding to a researcher's suspicion. The company has expanded into wet-lab work, operating a lab that conducts biology experiments.

The case shows both promise and the standard of evidence needed. Searching vast genomic databases in under a day is a real acceleration, and finding candidates for further study is valuable. Calling something a discovery, however, typically requires showing what a system does, with replicated experiments and peer review. Prior work by human researchers also shapes how much is new.

Biology is also an area where Anthropic has faced safety questions, including reports that users found ways around safeguards for bioweapons research. Peer-reviewed results describing the ART system's function, independent replication and clearer accounts of human versus AI contributions will determine how the claim is remembered. More in our research coverage and healthcare coverage.